initial release

.gitattributes

@@ -0,0 +1,38 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text
+ukisai-banner.png filter=lfs diff=lfs merge=lfs -text
+swift-1.5-planet-demo.mp4 filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,233 @@
+Swift Open License v1.0
+
+TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+1. Definitions.
+
+"License" shall mean the terms and conditions for use, reproduction, and
+distribution as defined by this document.
+
+"Licensor" shall mean UkisAI.
+
+"Legal Entity" shall mean the union of the acting entity and all other entities
+that control, are controlled by, or are under common control with that entity.
+For the purposes of this definition, "control" means (i) the power, direct or
+indirect, to cause the direction or management of such entity, whether by
+contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the
+outstanding shares, or (iii) beneficial ownership of such entity.
+
+"You" (or "Your") shall mean an individual or Legal Entity exercising
+permissions granted by this License.
+
+"Source" form shall mean the preferred form for making modifications,
+including but not limited to software source code, documentation source,
+configuration files, and model weights in an unquantized, trainable format.
+
+"Object" form shall mean any form resulting from mechanical transformation or
+translation of a Source form, including but not limited to compiled object
+code, generated documentation, quantized or otherwise converted model weights,
+and conversions to other media types or file formats.
+
+"Base Model" shall mean the Qwen3.8-27B model, Copyright 2026 Alibaba Cloud,
+made available at https://huggingface.co/Qwen/Qwen3.8-27B, including its
+weights, configuration, tokenizer, and chat template, in any form.
+
+"Base Model License" shall mean the Apache License, Version 2.0, under which
+the Base Model is made available. A copy is distributed with the Work in the
+file LICENSE-APACHE-2.0.
+
+"Swift Contribution" shall mean the modifications to the Base Model authored
+by Licensor, in any form, including without limitation adapted model weights,
+weight deltas, model weights to the extent they differ from the Base Model, and
+any configuration, documentation, and evaluation materials created by Licensor
+and distributed with the Work.
+
+"Work" shall mean the Swift Contribution together with, to the extent of
+Licensor's rights therein, the Derivative Work of the Base Model made available
+by Licensor under this License (as indicated by a copyright notice that is
+included in or attached to the work), in any format made available by Licensor.
+
+"Derivative Works" shall mean any work, whether in Source or Object form, that
+is based on (or derived from) the Work and for which the editorial revisions,
+annotations, elaborations, or other modifications represent, as a whole, an
+original work of authorship. For the purposes of this License, Derivative Works
+shall not include works that remain separable from, or merely link (or bind by
+name) to the interfaces of, the Work and Derivative Works thereof.
+
+"Contribution" shall mean any work of authorship, including the original
+version of the Work and any modifications or additions to that Work or
+Derivative Works thereof, that is intentionally submitted to Licensor for
+inclusion in the Work by the copyright owner or by an individual or Legal
+Entity authorized to submit on behalf of the copyright owner. For the purposes
+of this definition, "submitted" means any form of electronic, verbal, or
+written communication sent to the Licensor or its representatives, including
+but not limited to communication on electronic mailing lists, source code
+control systems, and issue tracking systems that are managed by, or on behalf
+of, the Licensor for the purpose of discussing and improving the Work, but
+excluding communication that is conspicuously marked or otherwise designated in
+writing by the copyright owner as "Not a Contribution."
+
+"Contributor" shall mean Licensor and any individual or Legal Entity on behalf
+of whom a Contribution has been received by Licensor and subsequently
+incorporated within the Work.
+
+"Commercial Use" shall mean any use of the Work or a Derivative Work for direct
+or indirect commercial advantage or monetary compensation.
+
+"Qualified Non-Profit Organization" shall mean a Legal Entity that is organized
+and operated exclusively for religious, charitable, scientific, testing for
+public safety, literary, or educational purposes, and which is exempt from
+federal income tax under Section 501(c)(3) of the United States Internal
+Revenue Code of 1986, as amended, or any equivalent non-profit or charitable
+organization in a foreign jurisdiction.
+
+"Non-Commercial or Research Purposes" shall mean purposes that do not involve
+any use of the Work or a Derivative Work for Commercial Use.
+
+"Threshold" shall mean gross revenue of one million United States dollars
+(US$1,000,000) or more, measured over the most recently completed fiscal year
+of You together with every Legal Entity that controls, is controlled by, or is
+under common control with You.
+
+2. Grant of Copyright License. Subject to the terms and conditions of this
+License, including the Commercial Use limitation set forth in Section 5, each
+Contributor hereby grants to You a perpetual, worldwide, non-exclusive,
+no-charge, royalty-free, irrevocable copyright license to reproduce, prepare
+Derivative Works of, publicly display, publicly perform, sublicense, and
+distribute the Work and such Derivative Works in Source or Object form.
+
+3. Grant of Patent License. Subject to the terms and conditions of this
+License, including the Commercial Use limitation set forth in Section 5, each
+Contributor hereby grants to You a perpetual, worldwide, non-exclusive,
+no-charge, royalty-free, irrevocable (except as stated in this section) patent
+license to make, have made, use, offer to sell, sell, import, and otherwise
+transfer the Work, where such license applies only to those patent claims
+licensable by such Contributor that are necessarily infringed by their
+Contribution(s) alone or by combination of their Contribution(s) with the Work
+to which such Contribution(s) was submitted. If You institute patent litigation
+against any entity (including a cross-claim or counterclaim in a lawsuit)
+alleging that the Work or a Contribution incorporated within the Work
+constitutes direct or contributory patent infringement, then any patent
+licenses granted to You under this License for that Work shall terminate as of
+the date such litigation is filed.
+
+4. Redistribution. You may reproduce and distribute copies of the Work or
+Derivative Works thereof in any medium, with or without modifications, and in
+Source or Object form, provided that You meet the following conditions:
+
+(a) You must give any other recipients of the Work or Derivative Works a copy
+of this License; and
+
+(b) You must cause any modified files to carry prominent notices stating that
+You changed the files; and
+
+(c) You must retain, in the Source form of any Derivative Works that You
+distribute, all copyright, patent, trademark, and attribution notices from the
+Source form of the Work, excluding those notices that do not pertain to any
+part of the Derivative Works; and
+
+(d) If the Work includes a "NOTICE" text file as part of its distribution, then
+any Derivative Works that You distribute must include a readable copy of the
+attribution notices contained within such NOTICE file, excluding those notices
+that do not pertain to any part of the Derivative Works, in at least one of the
+following places: within a NOTICE text file distributed as part of the
+Derivative Works; within the Source form or documentation, if provided along
+with the Derivative Works; or, within a display generated by the Derivative
+Works, if and wherever such third-party notices normally appear. The contents
+of the NOTICE file are for informational purposes only and do not modify the
+License. You may add Your own attribution notices within Derivative Works that
+You distribute, alongside or as an addendum to the NOTICE text from the Work,
+provided that such additional attribution notices cannot be construed as
+modifying the License; and
+
+(e) If the copy You distribute contains any portion of the Base Model
+(including merged, quantized, or otherwise converted weights that incorporate
+the Base Model), You must also give recipients a copy of the Base Model License
+and must comply with the Base Model License with respect to the Base Model.
+
+You may add Your own copyright statement to Your modifications and may provide
+additional or different license terms and conditions for use, reproduction, or
+distribution of Your modifications, or for any such Derivative Works as a
+whole, provided Your use, reproduction, and distribution of the Work otherwise
+complies with the conditions stated in this License, and provided that Section
+5 continues to apply to the Swift Contribution contained in any such
+Derivative Works.
+
+5. Commercial Use Limitation.
+
+(a) The rights granted under this License for Commercial Use are conditioned
+upon You or Your Legal Entity not exceeding the Threshold.
+
+(b) Any Commercial Use of the Work or a Derivative Work by a Legal Entity that
+exceeds the Threshold is not licensed under this License.
+
+(c) The Threshold shall not apply to a Qualified Non-Profit Organization's use
+of the Work or a Derivative Work for Non-Commercial or Research Purposes.
+
+(d) A Legal Entity that exceeds the Threshold may obtain a separate written
+license for Commercial Use from Licensor (the "Swift Enterprise License").
+Contact: https://ukisai.com/contact.
+
+6. Base Model Rights. The Work incorporates the Base Model. Nothing in this
+License limits, restricts, conditions, or modifies any rights You have in the
+Base Model under the Base Model License, and the Base Model remains available
+to You from its licensor under the Base Model License. Sections 5 and 12 of
+this License apply solely to the Swift Contribution and to the Work or any
+Derivative Work to the extent it contains, incorporates, or is derived from the
+Swift Contribution.
+
+7. Submission of Contributions. Unless You explicitly state otherwise, any
+Contribution intentionally submitted for inclusion in the Work by You to the
+Licensor shall be under the terms and conditions of this License, without any
+additional terms or conditions. Notwithstanding the above, nothing herein shall
+supersede or modify the terms of any separate license agreement you may have
+executed with Licensor regarding such Contributions.
+
+8. Trademarks. This License does not grant permission to use the trade names,
+trademarks, service marks, or product names of the Licensor (including "UkisAI"
+and "Swift"), except for the reasonable and customary use in describing the
+origin of the Work and reproducing the content of the NOTICE file.
+
+9. Disclaimer of Warranty. Unless required by applicable law or agreed to in
+writing, Licensor provides the Work (and each Contributor provides its
+Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied, including, without limitation, any warranties
+or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+PARTICULAR PURPOSE. You are solely responsible for determining the
+appropriateness of using or redistributing the Work and assume any risks
+associated with Your exercise of permissions under this License.
+
+10. Limitation of Liability. In no event and under no legal theory, whether in
+tort (including negligence), contract, or otherwise, unless required by
+applicable law (such as deliberate and grossly negligent acts) or agreed to in
+writing, shall any Contributor be liable to You for damages, including any
+direct, indirect, special, incidental, or consequential damages of any
+character arising as a result of this License or out of the use or inability to
+use the Work (including but not limited to damages for loss of goodwill, work
+stoppage, computer failure or malfunction, or any and all other commercial
+damages or losses), even if such Contributor has been advised of the
+possibility of such damages.
+
+11. Accepting Warranty or Additional Liability. While redistributing the Work
+or Derivative Works thereof, You may choose to offer, and charge a fee for,
+acceptance of support, warranty, indemnity, or other liability obligations
+and/or rights consistent with this License. However, in accepting such
+obligations, You may act only on Your own behalf and on Your sole
+responsibility, not on behalf of any other Contributor, and only if You agree
+to indemnify, defend, and hold each Contributor harmless for any liability
+incurred by, or claims asserted against, such Contributor by reason of your
+accepting any such warranty or additional liability.
+
+12. Termination. This License will terminate automatically and immediately if
+You fail to comply with any of its terms and conditions. Upon termination, You
+must cease all use of the Swift Contribution and of any Work or Derivative
+Works containing it, and delete all copies in Your possession. Termination does
+not affect Your rights in the Base Model under the Base Model License.
+
+END OF TERMS AND CONDITIONS
+
+APPENDIX: Notice for redistributors (quantizations, conversions, merges).
+
+   Copyright 2026 UkisAI. Swift Contribution licensed under the Swift Open
+   License v1.0 (https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b/blob/main/LICENSE).
+   Derivative of Qwen3.8-27B, Copyright 2026 Alibaba Cloud, Apache License 2.0.

LICENSE-APACHE-2.0

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Alibaba Cloud
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

NOTICE

@@ -0,0 +1,24 @@
+Swift 1.5 Qwen3.8-27B
+Copyright 2026 UkisAI
+
+UkisAI's contribution (the "Swift Contribution") is licensed under the
+Swift Open License v1.0. See LICENSE.
+
+This model is a Derivative Work of Qwen3.8-27B
+  https://huggingface.co/Qwen/Qwen3.8-27B
+  Copyright 2026 Alibaba Cloud
+  Licensed under the Apache License, Version 2.0. See LICENSE-APACHE-2.0.
+
+Swift 1.5 continues UkisAI's Swift 1.0 model, itself a Qwen3.8-27B
+derivative. The full release incorporates the Qwen3.8-27B Base Model and
+UkisAI's Swift 1.0 and Swift 1.5 contributions.
+
+Changes made by UkisAI (Apache License 2.0, Section 4(b) change notice):
+  - model-*.safetensors, model.safetensors.index.json: the model weights from
+    Swift 1.0 were further adapted by UkisAI using additional post-training
+    methods.
+  - README.md: replaced. LICENSE and NOTICE added.
+  - All other files (config.json, generation_config.json, chat_template.jinja,
+    tokenizer.json, tokenizer_config.json, vocab.json, merges.txt,
+    preprocessor_config.json, video_preprocessor_config.json) are retained from
+    the parent model and remain available under Apache License, Version 2.0.

QUANTIZATION_MANIFEST.json

@@ -0,0 +1,95 @@
+{
+  "checkpoint": {
+    "exact_unquantized_tensors": 609,
+    "quantized_modules": 590,
+    "saved_tensors": 2379,
+    "tensor_bytes": 19281804384,
+    "weight_shard_names": [
+      "model-00001-of-00004.safetensors",
+      "model-00002-of-00004.safetensors",
+      "model-00003-of-00004.safetensors",
+      "model-00004-of-00004.safetensors"
+    ],
+    "weight_shards": 4
+  },
+  "limitations": {
+    "broad_accuracy_benchmark": "NOT_RUN",
+    "full_apple_silicon_generation": "NOT_RUN",
+    "integrated_mtp_speculative_generation": "NOT_IMPLEMENTED_BY_PATCH",
+    "integrated_multimodal_generation": "NOT_IMPLEMENTED_BY_PATCH",
+    "long_context_benchmark": "NOT_RUN"
+  },
+  "mlx_lm_revision": "c69d1288440a0dc4e6401fc417098b07598dccd5",
+  "patches": [
+    {
+      "path": "compatibility/swift15-mlx-lm.patch",
+      "sha256": "f6f1d0bdafa45863bfbf93dac0398c481c993ea04fdf38b9bae98c643f89eaec"
+    },
+    {
+      "path": "compatibility/enable-5bit.patch",
+      "sha256": "b985961eac3035e05ca4c9f3a8b283c26e6dd69bab4d113997fc04fe2a4f99cc"
+    }
+  ],
+  "quantization": {
+    "bits": 5,
+    "group_size": 64,
+    "method": "official MLX-LM affine round-to-nearest conversion",
+    "mode": "affine"
+  },
+  "source_manifest_sha256": "0a00065b88ab003281853a7fb9bd5ce0086bc3781b36136d8c39da19933923ae",
+  "source_repository": "ukisai/Swift-1.5-Qwen3.8-27b",
+  "source_revision": "5ad04445d2686f525e9fbe5c077e6fa0c7df4200",
+  "source_revision_url": "https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b/tree/5ad04445d2686f525e9fbe5c077e6fa0c7df4200",
+  "source_tensors": 1199,
+  "source_weight_bytes": 55563006776,
+  "source_weight_shards": 18,
+  "status": "VALIDATED",
+  "validation": {
+    "all_floating_tensors_finite": true,
+    "all_source_tensors_accounted_for": 1199,
+    "load_memory_bytes": 19281804392,
+    "mtp": {
+      "path": "Explicit MTP step with real text hidden states and shared LM head; speculative generation is not integrated",
+      "shape": [
+        1,
+        1,
+        248320
+      ],
+      "status": "PASS"
+    },
+    "process_peak_rss_bytes": 24964259840,
+    "status": "PASS",
+    "text_generation": {
+      "elapsed_seconds": 495.4394222159999,
+      "finish_reason": "stop",
+      "prompt": "<|im_start|>user\nReply with exactly: Hello from Swift.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n",
+      "prompt_tokens_per_second": 0.04886516458777248,
+      "text": "Hello from Swift.",
+      "token_ids": [
+        9419,
+        494,
+        22929,
+        13,
+        248046
+      ],
+      "tokens": 5,
+      "tokens_per_second": 0.05894060179705689
+    },
+    "total_validation_seconds": 609.2779944080003,
+    "vision": {
+      "grid": [
+        [
+          1,
+          16,
+          16
+        ]
+      ],
+      "path": "Vision encoder only; image/video insertion and multimodal text generation are not implemented",
+      "shape": [
+        64,
+        5120
+      ],
+      "status": "PASS"
+    }
+  }
+}

README.md

@@ -0,0 +1,135 @@
+---
+license: other
+license_name: swift-open-license-1.0
+license_link: https://huggingface.co/ukisai/Swift-1.5-5bit-MLX/blob/main/LICENSE
+base_model: ukisai/Swift-1.5-Qwen3.8-27b
+base_model_relation: quantized
+library_name: mlx
+pipeline_tag: text-generation
+tags:
+  - mlx
+  - quantized
+  - 5-bit
+  - affine
+  - qwen3_8
+---
+
+<div align="center">
+  <a href="https://ukisai.com"><img src="ukisai-banner.png" alt="UkisAI" style="width:100%;max-width:100%;height:auto;display:block;margin-bottom:0.6em;" /></a>
+  <a href="https://ukisai.com">Website</a> &bull;
+  <a href="https://ukisai.com/products/swift">Learn more</a> &bull;
+  <a href="https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b">BF16 model</a> &bull;
+  <a href="https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27B-GGUF">GGUF</a> &bull;
+  <a href="https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27B-GSQ-RCO-GGUF">GSQ-RCO GGUF</a> &bull;
+  <a href="#evaluation">Evaluation</a> &bull;
+  <a href="#license-and-access">Enterprise licensing</a>
+</div>
+
+# Swift 1.5 Qwen3.8-27B — 5-bit MLX
+
+**MLX affine 5-bit quantization, group size 64.** Swift 1.5 is UkisAI's
+reasoning-efficient Qwen3.8-27B derivative, focused on long-horizon, agentic and
+coding tasks. This export preserves the text, vision and MTP parameter tree;
+its supported generation interface is text-only with the included MLX-LM patches.
+
+Swift 1.5 uses **58.5% fewer thinking tokens** than base Qwen3.8-27B while scoring **0.35% higher**, for a **9.18× speed-up** on several tasks.
+
+> [!CAUTION]
+> Use only a complete snapshot whose files match `UPLOAD_MANIFEST.json`.
+> Historical build tests are not certification of an incomplete Hub snapshot.
+> Full independent Apple Silicon generation and quality evaluation remain **NOT_RUN**.
+> The 19.28 GB tensor payload must not be forced onto a 16 GiB Mac.
+
+## Demo
+
+We gave base Qwen3.8-27B and Swift 1.5 27B the same prompt:
+
+> create a 3d little planet globe where I (player can walk around) and it has all these biomes to explore, the globe doesn't have to be too big, but still fun to go around. It's about a boy scout who is camping and goes around exploring.
+
+<video src="https://huggingface.co/ukisai/Swift-1.5-5bit-MLX/resolve/main/swift-1.5-planet-demo.mp4" controls autoplay muted loop playsinline style="width:100%;height:auto;border-radius:12px;"></video>
+
+Try the game yourself here: [https://ukisai.com/swift-games/27b](https://ukisai.com/swift-games/27b)
+
+Base Qwen3.8-27B took 104.6 minutes to build its game. Swift 1.5 took 11.39 minutes.
+
+## Source and quantization
+
+The recorded source is the complete customized Swift BF16 export at
+[`5ad04445d2686f525e9fbe5c077e6fa0c7df4200`](https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b/tree/5ad04445d2686f525e9fbe5c077e6fa0c7df4200),
+not base Qwen or another quantized model. The converter uses official MLX-LM
+commit `c69d1288440a0dc4e6401fc417098b07598dccd5` with the included architecture
+patch followed by the 5-bit extension.
+
+The original build report accounts for 1,199 source tensors, including 333 vision
+and 15 MTP tensors, and records 590 quantized modules plus 609 unquantized BF16
+tensors after layout mapping. Its four shards contain 2,379 saved tensors and
+19,281,804,384 bytes of tensor data. Independent recovery checks verified full
+SHA-256 hashes and header/index consistency of the original build. They did not
+repeat the complete source-value equality, finite-value or model-generation tests.
+
+The original tokenizer, template, configs, processors and quantized bytes are
+preserved. [QUANTIZATION_MANIFEST.json](QUANTIZATION_MANIFEST.json) and the
+existing `compatibility/` reports are historical build evidence. The current
+[upload manifest](UPLOAD_MANIFEST.json) identifies the intended complete file set.
+
+## Evaluation
+
+See the [Swift BF16 source evaluation](https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b#evaluation)
+for source benchmarks and methodology. They were not rerun on this MLX export.
+No new broad accuracy, stability, long-context or BF16 quality-parity result is claimed.
+
+## Validation and use
+
+**Both supplied MLX-LM patches are required, in order.** [USAGE.md](USAGE.md)
+contains the pinned install, full-file integrity check and generation example.
+The preserved source architecture is `Qwen3_5ForConditionalGeneration` / `qwen3_5`.
+
+The historical Linux build reports strict reload of 2,379 tensors, finite floating
+values, exact unquantized BF16 preservation, processor/tokenizer loading, and a short
+CPU generation returning `Hello from Swift.`. It also reports a real-weight vision
+encoder check and one explicit MTP step. These are historical results, not new
+independent inference results for the uploaded release.
+
+The CPU example promotes in-memory floating values to FP32 while retaining packed
+5-bit UINT32 weights. A new synthetic macOS CPU/Metal diagnostic reproduces an MLX
+0.32.2 CPU BF16 accumulation issue; it is not a Linux or full-model test. See
+[diagnostic](compatibility/macos-quantized-matmul-diagnostic.json) and
+[package checks](compatibility/package-checks.json).
+
+Full 27B Apple Silicon generation is unverified. Integrated image/video chat and
+speculative MTP generation are **not implemented** by the patch. Vision/MTP weights
+and component checks do not establish those end-to-end capabilities. Runtime/cache
+and OS memory must be budgeted in addition to the tensor payload.
+
+## License and access
+
+Swift 1.5 derives from [Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B)
+(Copyright 2026 Alibaba Cloud, [Apache License 2.0](https://huggingface.co/ukisai/Swift-1.5-5bit-MLX/blob/main/LICENSE-APACHE-2.0)).
+UkisAI's adapted weights are licensed under the [Swift Open License v1.0](https://huggingface.co/ukisai/Swift-1.5-5bit-MLX/blob/main/LICENSE).
+See [NOTICE](https://huggingface.co/ukisai/Swift-1.5-5bit-MLX/blob/main/NOTICE) for attribution and change notices.
+
+Personal, research, educational, evaluation and commercial use are free for
+individuals and organizations with gross annual revenue, including affiliates,
+of up to US$1,000,000. Above that threshold, commercial use requires a separate
+Swift Enterprise License. Contact [UkisAI](https://ukisai.com/contact) for terms.
+Nothing in the Swift Open License limits the Apache 2.0 rights in Qwen3.8-27B itself.
+The accompanying Apple MLX-LM code has a separate upstream
+[MIT notice](compatibility/LICENSE-MLX-LM-MIT).
+
+## Citation
+
+```bibtex
+@misc{swift-1.5-qwen3.8-27b,
+  title  = {Swift 1.5 Qwen3.8-27B},
+  author = {UkisAI},
+  year   = {2026},
+  url    = {https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b}
+}
+```
+
+## Acknowledgements
+
+We acknowledge the [NVIDIA Innovation Lab](https://www.nvidia.com/en-us/data-center/innovation-lab/),
+[Amazon Web Services](https://aws.amazon.com/), and [Google Cloud](https://cloud.google.com/)
+for compute credits and infrastructure support for Swift's development, training
+and evaluation.

UPLOAD_MANIFEST.json

@@ -0,0 +1,210 @@
+{
+  "files": [
+    {
+      "path": ".gitattributes",
+      "bytes": 1686,
+      "sha256": "8a9fd1f95733fb82a6cba50f5ea2b1e26ca7578bea7a7cc21b4301ff83b1ce3b",
+      "git_blob_sha1": "3727b14f485a05787b6cda82f0346eaeaa2e751b"
+    },
+    {
+      "path": "LICENSE",
+      "bytes": 13306,
+      "sha256": "1367057bf17041aa1d69286a1400be5f464d2f8b8f54f600088b209eac1850be",
+      "git_blob_sha1": "209a5720f7e4a747e658b49808a933309eeecceb"
+    },
+    {
+      "path": "LICENSE-APACHE-2.0",
+      "bytes": 11544,
+      "sha256": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a",
+      "git_blob_sha1": "f938136e3adacfd92be087f6e113b5d6d97f678f"
+    },
+    {
+      "path": "NOTICE",
+      "bytes": 1133,
+      "sha256": "be30f3d464974990e40e9833bc6f356fd89fe16733c6c1a581d97106b8ae741d",
+      "git_blob_sha1": "c4ad1a7188a4191cb5efd0f5109bf6e9ba8caee8"
+    },
+    {
+      "path": "QUANTIZATION_MANIFEST.json",
+      "bytes": 2946,
+      "sha256": "7bbc577963403dd60e93e994a8aeb12142558031d985489aea27708a97ada4cf",
+      "git_blob_sha1": "b25e2206cf43c5b85231da234a5014e1ea2e1e99"
+    },
+    {
+      "path": "README.md",
+      "bytes": 6283,
+      "sha256": "77bebdc893764e1ddb8499b4dd614856a33a39689ca03c46623c11f7a6480722",
+      "git_blob_sha1": "91d0f0da8a67b2a793ccb1fc16e54d6e14b67729"
+    },
+    {
+      "path": "USAGE.md",
+      "bytes": 4963,
+      "sha256": "8d9aeb9c1a86a8e23bc4daf4c7b38b3f8af7a7e0b8b74616e411346469b3cc0e",
+      "git_blob_sha1": "816e5c09652657535d15d4d8fddfb59b2253cd43"
+    },
+    {
+      "path": "chat_template.jinja",
+      "bytes": 8952,
+      "sha256": "c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041",
+      "git_blob_sha1": "c0c686f9c38d70d179fb7b5f5aa7530bc913dda3"
+    },
+    {
+      "path": "compatibility/LICENSE-MLX-LM-MIT",
+      "bytes": 1066,
+      "sha256": "ccfab7ccb2ea306f71531c8ca77bb55507606cd90768b1e32b8b52ab5b48cf01",
+      "git_blob_sha1": "98ff47b9ef9d4ac9f1a4bde3db13dd27456c5ea1"
+    },
+    {
+      "path": "compatibility/conversion-result.json",
+      "bytes": 694,
+      "sha256": "46d5f5abfae73b29c2b65ec5a6d05decd8e4d5b50eb0dd0df90f94edae040986",
+      "git_blob_sha1": "7b9848cd983f8f7df81116454907e733543fed82"
+    },
+    {
+      "path": "compatibility/enable-5bit.patch",
+      "bytes": 2227,
+      "sha256": "b985961eac3035e05ca4c9f3a8b283c26e6dd69bab4d113997fc04fe2a4f99cc",
+      "git_blob_sha1": "c577d5a2b6ea8fe634076ebd1387ecf7575d5d27"
+    },
+    {
+      "path": "compatibility/environment-linux.json",
+      "bytes": 274,
+      "sha256": "29646417ed15e6c1d08602b5832287abd44aabf4da624ea06f7e0b617bccbb48",
+      "git_blob_sha1": "0bca526431ce202cd367b40381cbdd041e02028e"
+    },
+    {
+      "path": "compatibility/macos-quantized-matmul-diagnostic.json",
+      "bytes": 4399,
+      "sha256": "90b992aa7535dde1ecbdcca93d6c8df78397689f4456e095abbae7c112d1754c",
+      "git_blob_sha1": "d9475a8bb176a03cf7e3b6fd686e56b59e93e297"
+    },
+    {
+      "path": "compatibility/package-checks.json",
+      "bytes": 77429,
+      "sha256": "016ebd02e676cdecebb27c9049bd53bff0f803a219b81ad364ca1d769d7bedc8",
+      "git_blob_sha1": "599f4e1d3b0e3c2d8032bc68d67b46a279490983"
+    },
+    {
+      "path": "compatibility/patch-manifest.json",
+      "bytes": 380,
+      "sha256": "c862249f212d4f94113a85336db22a24f3128014c034cdf9878fb9e44b103a95",
+      "git_blob_sha1": "955f2d916e47bab3292448d9db7681808053df02"
+    },
+    {
+      "path": "compatibility/quant-validation-results.json",
+      "bytes": 3784,
+      "sha256": "5a2bdbaa7ffa5616b2edf4817953369c694596c6d41aca2ba458a48b651ebab2",
+      "git_blob_sha1": "89c26ee2aabe9ce7717a0d91b217717af4d27b3f"
+    },
+    {
+      "path": "compatibility/source-structural-results.json",
+      "bytes": 2503,
+      "sha256": "cb9ccedd803a6cd336e7412de3da6809f89c30e567d3300e1c14af7f90e05fbf",
+      "git_blob_sha1": "b2641ec35a9f4810da73f8d439ec4603147d15a8"
+    },
+    {
+      "path": "compatibility/source-verification.json",
+      "bytes": 380649,
+      "sha256": "27d87d8d5209c6559643efdd7d7b468b0138ebfb3a7d5771b29be79214bffe75",
+      "git_blob_sha1": "c807fa472c01946b5ad47e1438f57e5980f01d38"
+    },
+    {
+      "path": "compatibility/swift15-mlx-lm.patch",
+      "bytes": 38976,
+      "sha256": "f6f1d0bdafa45863bfbf93dac0398c481c993ea04fdf38b9bae98c643f89eaec",
+      "git_blob_sha1": "b739acc2e3b4cb82f073b7280255bcb898bb2a33"
+    },
+    {
+      "path": "config.json",
+      "bytes": 4577,
+      "sha256": "77eaf6b1ee9e2ca06a9a3119f218853ec16c6f9f558e6b492a7e9a1be8d2ac43",
+      "git_blob_sha1": "3b9f5cd48042e788592ec30c3e153ffe794e45c7"
+    },
+    {
+      "path": "generation_config.json",
+      "bytes": 202,
+      "sha256": "e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e",
+      "git_blob_sha1": "023756cfadf88e5bf69eefeee3e172f38c448d64"
+    },
+    {
+      "path": "merges.txt",
+      "bytes": 3353259,
+      "sha256": "a9d356d7bdf1ef4949e3e748e95b8e10ad9d4e2e838eddc38a0a7b6b94d1db8d",
+      "git_blob_sha1": "a494e019ca1502219fd0128658b979e5f05ae8e8"
+    },
+    {
+      "path": "model-00001-of-00004.safetensors",
+      "bytes": 5360417859,
+      "sha256": "9da8750595ec42d8c5d2ff0ae0b695e387cbcb3c0beac40f262e6d9a7196b59d"
+    },
+    {
+      "path": "model-00002-of-00004.safetensors",
+      "bytes": 5338181777,
+      "sha256": "b592723bd4e9f01344335cd7d4b4f41a7c8cde98e2e074ad1d943cdb951a6835"
+    },
+    {
+      "path": "model-00003-of-00004.safetensors",
+      "bytes": 5355925620,
+      "sha256": "0027e7fe05310d5b2d960414221ed0872de9b511aa7e6b1c3a8f0e4647558412"
+    },
+    {
+      "path": "model-00004-of-00004.safetensors",
+      "bytes": 3227578201,
+      "sha256": "afec577f16805b327471561d6841662eb4acf7d866ad5e761c5ef8cd3149d8f4"
+    },
+    {
+      "path": "model.safetensors.index.json",
+      "bytes": 232537,
+      "sha256": "83526f2e7856d41145f3599146a4a574ea2b258c4abe09302617243f13abec62",
+      "git_blob_sha1": "936e7dc269c4155002581e4c5620f085ab1d64f0"
+    },
+    {
+      "path": "preprocessor_config.json",
+      "bytes": 390,
+      "sha256": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516",
+      "git_blob_sha1": "2ea84a437d448ff71b08df68fdd949d5cc4ebb64"
+    },
+    {
+      "path": "swift-1.5-planet-demo.mp4",
+      "bytes": 4738825,
+      "sha256": "1cda6924169e8b83e5d6d7299baf8329ed1b1a182a481aa0e06344e0f5c00111"
+    },
+    {
+      "path": "tokenizer.json",
+      "bytes": 12809320,
+      "sha256": "0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3"
+    },
+    {
+      "path": "tokenizer_config.json",
+      "bytes": 17928,
+      "sha256": "b11349aafa7cdc6a320767cf7ceb29ed82f7eda5d65e8e0819e76f0ce947bf27",
+      "git_blob_sha1": "5de744b3fca2129d7186979ae47c06be33903243"
+    },
+    {
+      "path": "ukisai-banner.png",
+      "bytes": 307739,
+      "sha256": "8577252b8b37e331f06f7ddecdf3bfa1c497e164f9eb36d78c37232d1b8492ca"
+    },
+    {
+      "path": "verify_release.py",
+      "bytes": 6662,
+      "sha256": "02ac6ed751c4dd915d1673f2d501f3322d27039c3a32a5bc3c0264861e4d0982",
+      "git_blob_sha1": "7e235791f8cfadb8321321231b3ea6359565db89"
+    },
+    {
+      "path": "video_preprocessor_config.json",
+      "bytes": 385,
+      "sha256": "7768af27c1fafa9cc9011c1dc20067e03f8915e03b63504550e11d5066986d13",
+      "git_blob_sha1": "3ba673a5ad7d4d13f54155ecd38b2a94a6dac8fe"
+    },
+    {
+      "path": "vocab.json",
+      "bytes": 6722759,
+      "sha256": "ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003",
+      "git_blob_sha1": "0aa0ce0658d60ac4a5d609f4eadb0e8e43514176"
+    }
+  ],
+  "file_count": 35,
+  "total_bytes": 19310861234,
+  "self_excluded": true
+}

USAGE.md

@@ -0,0 +1,100 @@
+# Swift 1.5 5-bit — complete MLX architecture
+
+Use only a complete snapshot: all four shards, original index, tokenizer and
+processor/runtime files are required. The 19.28 GB tensor payload plus runtime,
+cache and OS must fit available memory. Do not load this full model on a 16 GiB Mac
+or raise system memory limits to conceal insufficient hardware.
+
+## Pin the snapshot and verify it
+
+Use a new working directory. After installing the HF CLI, run `hf auth login`
+interactively if not already signed in with access to this private repository.
+The command below resolves current main once to a full commit and then uses
+only that pinned snapshot. For a repeat run, reuse the recorded commit.
+Do not use an incomplete historical upload or proceed after verification failure.
+
+```bash
+python3.12 -m venv .venv-swift5
+source .venv-swift5/bin/activate
+python -m pip install 'huggingface_hub==1.31.0'
+SWIFT_MLX_REVISION="$(python -c 'from huggingface_hub import HfApi; print(HfApi().model_info("ukisai/Swift-1.5-5bit-MLX").sha)')"
+printf 'Pinned model revision: %s\n' "$SWIFT_MLX_REVISION"
+hf download ukisai/Swift-1.5-5bit-MLX --revision "$SWIFT_MLX_REVISION" --local-dir Swift-1.5-5bit-MLX
+hf cache verify ukisai/Swift-1.5-5bit-MLX --revision "$SWIFT_MLX_REVISION" --local-dir Swift-1.5-5bit-MLX --fail-on-missing-files
+python Swift-1.5-5bit-MLX/verify_release.py Swift-1.5-5bit-MLX
+```
+
+The included checker has no network or model-loading code. Supply its optional
+`--manifest-sha256` argument from a trusted release plan to pin the manifest too.
+Without that trusted digest it checks consistency, not source authenticity.
+Stop if any file, checksum, index entry or payload-boundary check fails.
+
+## Install the patches in this order
+
+```bash
+git clone https://github.com/ml-explore/mlx-lm.git swift5-mlx-lm
+git -C swift5-mlx-lm checkout --detach c69d1288440a0dc4e6401fc417098b07598dccd5
+git -C swift5-mlx-lm apply --check ../Swift-1.5-5bit-MLX/compatibility/swift15-mlx-lm.patch
+git -C swift5-mlx-lm apply ../Swift-1.5-5bit-MLX/compatibility/swift15-mlx-lm.patch
+git -C swift5-mlx-lm apply --check ../Swift-1.5-5bit-MLX/compatibility/enable-5bit.patch
+git -C swift5-mlx-lm apply ../Swift-1.5-5bit-MLX/compatibility/enable-5bit.patch
+```
+
+Apple Silicon:
+
+```bash
+python -m pip install 'mlx==0.32.2' 'transformers==5.14.1' 'huggingface_hub==1.31.0' 'pillow==12.3.0'
+python -m pip install -e ./swift5-mlx-lm
+```
+
+Linux CPU, Python 3.12, glibc 2.35 or newer:
+
+```bash
+python -m pip install 'mlx[cpu]==0.32.2' 'transformers==5.14.1' 'huggingface_hub==1.31.0' 'pillow==12.3.0'
+python -m pip install -e ./swift5-mlx-lm
+```
+
+The historical [Linux environment](compatibility/environment-linux.json) records
+Hub 1.32.0. The 1.31.0 pin above was separately installed in the independent macOS
+audit, where 18 synthetic patch tests passed. These environments are not identical
+and the tests did not load the full 27B model. The upstream code
+[MIT notice](compatibility/LICENSE-MLX-LM-MIT) is included separately from weight licenses.
+
+## Text generation
+
+```python
+import mlx.core as mx
+from mlx_lm import generate, load
+from mlx_lm.sample_utils import make_sampler
+
+model, tokenizer = load("Swift-1.5-5bit-MLX")
+if mx.default_device() == mx.cpu:
+    model.apply(lambda x: x.astype(mx.float32) if mx.issubdtype(x.dtype, mx.floating) else x)
+prompt = tokenizer.apply_chat_template(
+    [{"role": "user", "content": "Reply with exactly: Hello from Swift."}],
+    tokenize=False, add_generation_prompt=True, enable_thinking=False,
+)
+mx.random.seed(20260922)
+print(generate(model, tokenizer, prompt=prompt, max_tokens=32, sampler=make_sampler(temp=0)))
+```
+
+The CPU branch changes only in-memory floating types; packed UINT32 weights and
+files are unchanged. The historical build reported a Linux BF16 accumulation issue.
+Its original diagnostic file was not published. The separately executed
+[macOS CPU/Metal diagnostic](compatibility/macos-quantized-matmul-diagnostic.json)
+uses synthetic tensors: summing 8,192 ones gives 256 on CPU BF16 and 8,192 with FP32;
+Metal BF16/FP32 also give 8,192. It is not a new Linux or full-model generation test.
+
+The template supports `reasoning_effort="low"`, `"medium"`, and `"xhigh"`.
+Template support does not establish generated quality for those modes.
+The explicit `model.mtp_logits` step and `model.visual` encoder have historical
+component evidence. Integrated image/video chat and speculative generation are
+not implemented; unsupported multimodal generation must not be reported as working.
+
+## Conversion provenance
+
+Conversion is not part of installation. If separately authorized, use only the
+complete customized Swift BF16 source identified in `QUANTIZATION_MANIFEST.json`,
+with its 18 shards verified before conversion, the pinned patched converter, and
+affine / 5-bit / group size 64. Do not fill missing weights with base Qwen or another
+quantization. Existing quantized weights are unchanged by these packaging repairs.

chat_template.jinja

@@ -0,0 +1,170 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- set reasoning_instructions = '' %}
+{%- if enable_thinking is undefined or enable_thinking is true %}
+    {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}
+    {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}
+        {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}
+    {%- endif %}
+    {%- if resolved_reasoning_effort == 'xhigh' %}
+        {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}
+    {%- elif resolved_reasoning_effort == 'low' %}
+        {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}
+    {%- endif %}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {%- if reasoning_instructions %}
+        {{- reasoning_instructions + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '<|im_start|>system\n' + (reasoning_instructions + '\n\n' if reasoning_instructions else '')  + content + '<|im_end|>\n' }}
+        {%- elif reasoning_instructions %}
+            {{- '<|im_start|>system\n' + reasoning_instructions + '<|im_end|>\n' }}
+        {%- endif %}
+    {%- elif reasoning_instructions %}
+        {{- '<|im_start|>system\n' + reasoning_instructions + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined and tool_call.arguments != '' %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

compatibility/LICENSE-MLX-LM-MIT

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright © 2023 Apple Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

compatibility/conversion-result.json

@@ -0,0 +1,28 @@
+{
+  "command": [
+    "mlx_lm.convert",
+    "--hf-path",
+    "<SOURCE_MODEL_DIR>",
+    "--mlx-path",
+    "<OUTPUT_MODEL_DIR>",
+    "--quantize",
+    "--q-mode",
+    "affine",
+    "--q-bits",
+    "5",
+    "--q-group-size",
+    "64"
+  ],
+  "elapsed_seconds": 130.01374627799942,
+  "finished_at": "2026-09-22T12:05:47.557129+00:00",
+  "quantization": {
+    "bits": 5,
+    "group_size": 64,
+    "mode": "affine"
+  },
+  "returncode": 0,
+  "source_manifest_sha256": "0a00065b88ab003281853a7fb9bd5ce0086bc3781b36136d8c39da19933923ae",
+  "source_repo": "ukisai/Swift-1.5-Qwen3.8-27b",
+  "source_revision": "5ad04445d2686f525e9fbe5c077e6fa0c7df4200",
+  "started_at": "2026-09-22T12:03:37.543218+00:00"
+}

compatibility/enable-5bit.patch

@@ -0,0 +1,47 @@
+diff --git a/mlx_lm/models/qwen3_5_full.py b/mlx_lm/models/qwen3_5_full.py
+--- a/mlx_lm/models/qwen3_5_full.py
++++ b/mlx_lm/models/qwen3_5_full.py
+@@ -459,9 +459,9 @@ class Model(nn.Module):
+                 settings.get("group_size"),
+                 settings.get("mode", "affine"),
+             )
+-            if (bits, group, mode) != (4, 64, "affine"):
++            if bits not in (4, 5) or group != 64 or mode != "affine":
+                 raise ValueError(
+-                    "This extension only prepares affine/4-bit/group-64 native checkpoints"
++                    "This extension only prepares affine/4-or-5-bit/group-64 native checkpoints"
+                 )
+             original = shapes[f"{path}.weight"]
+             if original[-1] % group:
+diff --git a/tests/test_qwen3_5_full.py b/tests/test_qwen3_5_full.py
+--- a/tests/test_qwen3_5_full.py
++++ b/tests/test_qwen3_5_full.py
+@@ -340,14 +340,17 @@ def test_official_nonquantized_convert_and_reload_preserve_all_tensors(
+         load_model(output, lazy=True)
+ 
+ 
+-def test_fixed_affine_quantized_roundtrip_preserves_component_tree(reference, tmp_path):
++@pytest.mark.parametrize("bits", [4, 5])
++def test_fixed_affine_quantized_roundtrip_preserves_component_tree(
++    reference, tmp_path, bits
++):
+     from mlx_lm.utils import quantize_model, save_model
+ 
+     config, _, _, state, _ = reference
+     model = Model(ModelArgs.from_dict(config))
+     model.load_weights(list(model.sanitize(state).items()), strict=True)
+     original_names = set(dict(tree_flatten(model.parameters())))
+-    model, quantized_config = quantize_model(model, config, 64, 4, mode="affine")
++    model, quantized_config = quantize_model(model, config, 64, bits, mode="affine")
+     save_model(tmp_path, model)
+     save_config(quantized_config, tmp_path / "config.json")
+     loaded, saved_config = load_model(tmp_path, lazy=False, strict=True)
+@@ -355,7 +358,7 @@ def test_fixed_affine_quantized_roundtrip_preserves_component_tree(reference, tm
+     assert set(actual) == set(expected)
+     assert original_names <= set(actual)
+     assert saved_config["quantization"] == {
+-        "bits": 4,
++        "bits": bits,
+         "group_size": 64,
+         "mode": "affine",
+     }

compatibility/environment-linux.json

@@ -0,0 +1,13 @@
+{
+  "packages": {
+    "huggingface-hub": "1.32.0",
+    "mlx": "0.32.2",
+    "mlx-cpu": "0.32.2",
+    "mlx-lm": "0.32.0",
+    "pillow": "12.3.0",
+    "safetensors": "0.8.0",
+    "transformers": "5.14.1"
+  },
+  "platform": "Linux x86_64",
+  "runtime": "CPU validation only"
+}

compatibility/macos-quantized-matmul-diagnostic.json

@@ -0,0 +1,189 @@
+{
+  "scope": "Synthetic kernel probe on macOS, not Linux and not real model inference",
+  "results": [
+    {
+      "device": "DeviceType.gpu",
+      "bits": 3,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.bfloat16",
+      "expected": 8192,
+      "actual": [
+        [
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0
+        ]
+      ],
+      "exact": true,
+      "packed_sha256": "de676bae28a480011d3d012db14bef539324e62a841a9627863c689bea168af3",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.02095266600008472
+    },
+    {
+      "device": "DeviceType.gpu",
+      "bits": 3,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.float32",
+      "expected": 8192,
+      "actual": [
+        [
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0
+        ]
+      ],
+      "exact": true,
+      "packed_sha256": "de676bae28a480011d3d012db14bef539324e62a841a9627863c689bea168af3",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.014540874999511288
+    },
+    {
+      "device": "DeviceType.gpu",
+      "bits": 5,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.bfloat16",
+      "expected": 8192,
+      "actual": [
+        [
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0
+        ]
+      ],
+      "exact": true,
+      "packed_sha256": "02b1c2234680617802901a77eae606ad02e4ddb4282ccbc60061eac5b2d90bba",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.017056000000593485
+    },
+    {
+      "device": "DeviceType.gpu",
+      "bits": 5,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.float32",
+      "expected": 8192,
+      "actual": [
+        [
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0
+        ]
+      ],
+      "exact": true,
+      "packed_sha256": "02b1c2234680617802901a77eae606ad02e4ddb4282ccbc60061eac5b2d90bba",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.021967042001051595
+    },
+    {
+      "device": "DeviceType.cpu",
+      "bits": 3,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.bfloat16",
+      "expected": 8192,
+      "actual": [
+        [
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0
+        ]
+      ],
+      "exact": false,
+      "packed_sha256": "de676bae28a480011d3d012db14bef539324e62a841a9627863c689bea168af3",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.00037379100103862584
+    },
+    {
+      "device": "DeviceType.cpu",
+      "bits": 3,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.float32",
+      "expected": 8192,
+      "actual": [
+        [
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0
+        ]
+      ],
+      "exact": true,
+      "packed_sha256": "de676bae28a480011d3d012db14bef539324e62a841a9627863c689bea168af3",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.00011829099821625277
+    },
+    {
+      "device": "DeviceType.cpu",
+      "bits": 5,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.bfloat16",
+      "expected": 8192,
+      "actual": [
+        [
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0,
+          256.0
+        ]
+      ],
+      "exact": false,
+      "packed_sha256": "02b1c2234680617802901a77eae606ad02e4ddb4282ccbc60061eac5b2d90bba",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 0.0003301249998912681
+    },
+    {
+      "device": "DeviceType.cpu",
+      "bits": 5,
+      "group_size": 64,
+      "floating_dtype": "mlx.core.float32",
+      "expected": 8192,
+      "actual": [
+        [
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0,
+          8192.0
+        ]
+      ],
+      "exact": true,
+      "packed_sha256": "02b1c2234680617802901a77eae606ad02e4ddb4282ccbc60061eac5b2d90bba",
+      "packed_dtype": "mlx.core.uint32",
+      "seconds": 8.200000229408033e-05
+    }
+  ]
+}

compatibility/package-checks.json

@@ -0,0 +1,2088 @@
+{
+  "observed_at": "2026-09-22T14:48:11.669625+00:00",
+  "source": "Independent local packaging audit",
+  "scope": "Original-build file integrity and runtime assets; no full model generation",
+  "original_build_integrity": {
+    "model": "mlx5",
+    "event": "SUMMARY",
+    "file_count": 29,
+    "tensor_count": 2379,
+    "manifest_sha256": "1e0d5730ef70b67f8dbbe6c2b2ced6115f74658b7a30c73948fa5497c226935a",
+    "manifest_mismatches": [],
+    "header_index_status": "PASS",
+    "elapsed_seconds": 147.257588171,
+    "scope": "Original build; NOT HF upload validation"
+  },
+  "original_small_file_matches": [
+    {
+      "file": "LICENSE",
+      "sha256": "1367057bf17041aa1d69286a1400be5f464d2f8b8f54f600088b209eac1850be",
+      "matches_original_build": true
+    },
+    {
+      "file": "LICENSE-APACHE-2.0",
+      "sha256": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a",
+      "matches_original_build": true
+    },
+    {
+      "file": "NOTICE",
+      "sha256": "be30f3d464974990e40e9833bc6f356fd89fe16733c6c1a581d97106b8ae741d",
+      "matches_original_build": true
+    },
+    {
+      "file": "QUANTIZATION_MANIFEST.json",
+      "sha256": "7bbc577963403dd60e93e994a8aeb12142558031d985489aea27708a97ada4cf",
+      "matches_original_build": true
+    },
+    {
+      "file": "README.md",
+      "sha256": "22ec5682ebce0fd3f576d70c87fc8bba81589ebdb6370f59bf664f36bf9bc694",
+      "matches_original_build": true
+    },
+    {
+      "file": "USAGE.md",
+      "sha256": "15e984ebae2115a54d86282ce7e04ebac6e78eb1e2ca4f55191d5a9a908b47a2",
+      "matches_original_build": true
+    },
+    {
+      "file": "chat_template.jinja",
+      "sha256": "c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/conversion-result.json",
+      "sha256": "46d5f5abfae73b29c2b65ec5a6d05decd8e4d5b50eb0dd0df90f94edae040986",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/enable-5bit.patch",
+      "sha256": "b985961eac3035e05ca4c9f3a8b283c26e6dd69bab4d113997fc04fe2a4f99cc",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/environment-linux.json",
+      "sha256": "29646417ed15e6c1d08602b5832287abd44aabf4da624ea06f7e0b617bccbb48",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/patch-manifest.json",
+      "sha256": "c862249f212d4f94113a85336db22a24f3128014c034cdf9878fb9e44b103a95",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/quant-validation-results.json",
+      "sha256": "5a2bdbaa7ffa5616b2edf4817953369c694596c6d41aca2ba458a48b651ebab2",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/source-structural-results.json",
+      "sha256": "2c07ba0d068c3cf8059b3cf31578c9e98540fb08140a2623d79a1eb9737e12e4",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/source-verification.json",
+      "sha256": "27d87d8d5209c6559643efdd7d7b468b0138ebfb3a7d5771b29be79214bffe75",
+      "matches_original_build": true
+    },
+    {
+      "file": "compatibility/swift15-mlx-lm.patch",
+      "sha256": "f6f1d0bdafa45863bfbf93dac0398c481c993ea04fdf38b9bae98c643f89eaec",
+      "matches_original_build": true
+    },
+    {
+      "file": "config.json",
+      "sha256": "77eaf6b1ee9e2ca06a9a3119f218853ec16c6f9f558e6b492a7e9a1be8d2ac43",
+      "matches_original_build": true
+    },
+    {
+      "file": "generation_config.json",
+      "sha256": "e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e",
+      "matches_original_build": true
+    },
+    {
+      "file": "merges.txt",
+      "sha256": "a9d356d7bdf1ef4949e3e748e95b8e10ad9d4e2e838eddc38a0a7b6b94d1db8d",
+      "matches_original_build": true
+    },
+    {
+      "file": "preprocessor_config.json",
+      "sha256": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516",
+      "matches_original_build": true
+    },
+    {
+      "file": "tokenizer.json",
+      "sha256": "0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3",
+      "matches_original_build": true
+    },
+    {
+      "file": "tokenizer_config.json",
+      "sha256": "b11349aafa7cdc6a320767cf7ceb29ed82f7eda5d65e8e0819e76f0ce947bf27",
+      "matches_original_build": true
+    },
+    {
+      "file": "ukisai-banner.png",
+      "sha256": "8577252b8b37e331f06f7ddecdf3bfa1c497e164f9eb36d78c37232d1b8492ca",
+      "matches_original_build": true
+    },
+    {
+      "file": "video_preprocessor_config.json",
+      "sha256": "7768af27c1fafa9cc9011c1dc20067e03f8915e03b63504550e11d5066986d13",
+      "matches_original_build": true
+    },
+    {
+      "file": "vocab.json",
+      "sha256": "ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003",
+      "matches_original_build": true
+    }
+  ],
+  "original_small_file_matches_scope": "Verified inputs before the documented documentation overlay; current file digests are in UPLOAD_MANIFEST.json.",
+  "tokenizer_comparison": {
+    "status": "PASS",
+    "cases": [
+      {
+        "case": "copy-01",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4c818d1bec319bdcb9f9ae6fb165958cc228993b88271d074717a54da892a12c",
+        "token_ids_sha256": "a92b1c7a4f0c7046a5d8adfa819d3d92870cce145b5316398b467f77267b467b"
+      },
+      {
+        "case": "copy-01",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "65d21c471e0404dd574585906ea708c25e143644e26d6f6da38a963c89d95353",
+        "token_ids_sha256": "2f2ea6c38bd950ee61cab76d0e0a2719b90793c129881f4653665f6f32e83f37"
+      },
+      {
+        "case": "copy-01",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "18150302804516ca9bde6d9f4859e35de3047beb7f3f42db2c709f9f659f7d97",
+        "token_ids_sha256": "99796b2eb8cb5ea634506794c13de258c6a4381ef2ff617a4952a2e81100a773"
+      },
+      {
+        "case": "copy-01",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "cc38501bdf707099ff6a0e337475ea7df68417bfc4012dbeff043f22d9d671ef",
+        "token_ids_sha256": "628de4373c94835dfb58149eb1f3a6cc7de44019ecff368687b61a6271501263"
+      },
+      {
+        "case": "copy-01",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4c818d1bec319bdcb9f9ae6fb165958cc228993b88271d074717a54da892a12c",
+        "token_ids_sha256": "a92b1c7a4f0c7046a5d8adfa819d3d92870cce145b5316398b467f77267b467b"
+      },
+      {
+        "case": "copy-02",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c899009f64185c1cf610481f00667851c613755d9992bbea2782f63a23aeab50",
+        "token_ids_sha256": "a2c44489d2864ff48ab7c3c4ed23786132dba628a7160d29d49538eb7754232d"
+      },
+      {
+        "case": "copy-02",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0dce1e6a9eae5b1630ac76c0d094e90b3d51252fe393b03eee6678564e29f43a",
+        "token_ids_sha256": "ca5fbed68bb860caa55ef73136f045a99917c48c8897883c60f8c71646f03e4f"
+      },
+      {
+        "case": "copy-02",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "398a9d6fee34ee4123220a2a2902f597f4313c8511d97f6793d6b546466c0421",
+        "token_ids_sha256": "d2a89478a168aa84c857b0a92a02e92d72827dafe25d4b295a85cb6e4ae62451"
+      },
+      {
+        "case": "copy-02",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "34206e0e86b7b48b6e4a90b432c00dfecd7e215379355ddf16616a891ec15a5f",
+        "token_ids_sha256": "f6d035e18f6cce3a81e364b17997f04a6e373fda9ffd2c803957a0b67a91f3b3"
+      },
+      {
+        "case": "copy-02",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c899009f64185c1cf610481f00667851c613755d9992bbea2782f63a23aeab50",
+        "token_ids_sha256": "a2c44489d2864ff48ab7c3c4ed23786132dba628a7160d29d49538eb7754232d"
+      },
+      {
+        "case": "copy-03",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7546d15c056af8aa4c0c90d23aabb5b1ca726609ba850c3b1b6185a613837c92",
+        "token_ids_sha256": "ddbc030acb515ce5e2d03db0a169b6f68905c4827af8a329fed19bec400f985f"
+      },
+      {
+        "case": "copy-03",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "941b6b40fb0afa04dcb3cbeb3d703e1362bfab8aff0c223fe723ad68fb0f1c8c",
+        "token_ids_sha256": "1ae80f7895f4296112ee802500c9650fbba344adbd43ea18f6ec38ee5fc1c5ab"
+      },
+      {
+        "case": "copy-03",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "fff8cb95ccca437354811f847c4803f4db097acb3e25741571dc0bc3125275d9",
+        "token_ids_sha256": "e2d2fbfdef377e1dc40cc011dd75adf8484f8ad6f2769241207f1070e4b4101f"
+      },
+      {
+        "case": "copy-03",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "b5b000a062e3a761951c5b319766cec522bfb6d922409fcabfc3d0ddb1a4359d",
+        "token_ids_sha256": "262b04aaeff8bf09922edc2a45504750fd9fc9c3f3055b403a9f68b871c6d1c9"
+      },
+      {
+        "case": "copy-03",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7546d15c056af8aa4c0c90d23aabb5b1ca726609ba850c3b1b6185a613837c92",
+        "token_ids_sha256": "ddbc030acb515ce5e2d03db0a169b6f68905c4827af8a329fed19bec400f985f"
+      },
+      {
+        "case": "copy-04",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "93cd159f2fb657d1616693847e4f5c43af88edb03234c0b2dcc0181ff408b6ed",
+        "token_ids_sha256": "9cca472a04965a233f1a5e45043c6f112b9ee441dd9b26aa23de8b4ca61a626c"
+      },
+      {
+        "case": "copy-04",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0d7641637466c7079c2bc3e9e79a92397c06d7622bb8472c475f0fe452654c83",
+        "token_ids_sha256": "f283875df94548c1657a5550db42828ce955ec3b20fa148e646bbf042f6f0d2b"
+      },
+      {
+        "case": "copy-04",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "05353d7dc1f61ff121bf6598e0936277bd9ca84fbc19fd91e286d6e4b45bc059",
+        "token_ids_sha256": "2f715ed2fb5ec71a98be5bd3ed990a1e70d0678f168da4306a5f27dddc0b6856"
+      },
+      {
+        "case": "copy-04",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "f1a18e28b7392715dc638885ae8d79244fbd278f43e7c13f69fe8fb2b93da608",
+        "token_ids_sha256": "433b7e2976986230f0a2e200c658f3a7f00f0e38fe8906970aaedddc66fa59d9"
+      },
+      {
+        "case": "copy-04",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "93cd159f2fb657d1616693847e4f5c43af88edb03234c0b2dcc0181ff408b6ed",
+        "token_ids_sha256": "9cca472a04965a233f1a5e45043c6f112b9ee441dd9b26aa23de8b4ca61a626c"
+      },
+      {
+        "case": "copy-05",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "fb61ba28f75ad7405020c6d3241f3728b010b8633fdc1642e129b81523e87ad3",
+        "token_ids_sha256": "c0337ac4d3e6f3afd6fb5313d1ae35c3fba5fdc1d3b32c615ea1a9d469b23426"
+      },
+      {
+        "case": "copy-05",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "a6bf42558b20071884725645d2ff092639c819386f92af80f95cf87951eceb25",
+        "token_ids_sha256": "c1f030f36157b04cdd7a5599f05578720ef484f3300982179014968be81ea065"
+      },
+      {
+        "case": "copy-05",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ef76dafe530370c7baa9a57c590899bfd62f6814b9c6c36735e28002e382a072",
+        "token_ids_sha256": "587c1d346c730fdf014d7e27e58f57344dc0d1e2c3a95a80dce943a16b5d21a6"
+      },
+      {
+        "case": "copy-05",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "96eaf3b174a72fcabbd9115c37f1a8c180a2ed19340a0cab757c80887a7b06dd",
+        "token_ids_sha256": "99598ba689281150a6a056699b14f24b499fc7c0a882ef22e4b2a2a682ff0057"
+      },
+      {
+        "case": "copy-05",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "fb61ba28f75ad7405020c6d3241f3728b010b8633fdc1642e129b81523e87ad3",
+        "token_ids_sha256": "c0337ac4d3e6f3afd6fb5313d1ae35c3fba5fdc1d3b32c615ea1a9d469b23426"
+      },
+      {
+        "case": "arithmetic-01",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "2a3e217d73ae1cdf4118b870a9ec322dc88de707843c83f0f5481f813faa99a4",
+        "token_ids_sha256": "56a79c0ee7bb6e52c284a19a16146c231142c47b22bfd0a27dac2330822afb49"
+      },
+      {
+        "case": "arithmetic-01",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ee9178c74035beb407c016586ef958578e5be4041b572aa2bf5409dcddf07739",
+        "token_ids_sha256": "b520feeac9ab433d35a057c9dc8e2a169254fa0db38e76955d042a2f91c774a7"
+      },
+      {
+        "case": "arithmetic-01",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "3c43b55f9ffffff2dcf76e9b6ae04a1311b330b2c0757ed1fe664cd64c469506",
+        "token_ids_sha256": "3b9b091b00ae6093fb726ab7fa7c5e5bdaf6aa87238eb14291bf52a439e178cb"
+      },
+      {
+        "case": "arithmetic-01",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "39e3c8c4b7c6b8352df4c942933b289007e0e9ac3c7d3dfa172203f0891ad801",
+        "token_ids_sha256": "17ca2fe9b641d0b785ef902f06a0fb4463f8b144e85903dcd0dbf7dea82f54be"
+      },
+      {
+        "case": "arithmetic-01",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "2a3e217d73ae1cdf4118b870a9ec322dc88de707843c83f0f5481f813faa99a4",
+        "token_ids_sha256": "56a79c0ee7bb6e52c284a19a16146c231142c47b22bfd0a27dac2330822afb49"
+      },
+      {
+        "case": "arithmetic-02",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "1580424514af0e654467483640add4ab4c970be8278db0a7dc51a6634160fafe",
+        "token_ids_sha256": "bc696d41dda4cfadb0539d3e6b16bba3012ff8adfd9eeba7f192b3fb020bd1ba"
+      },
+      {
+        "case": "arithmetic-02",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ed261285116d3895ef5f2209e9cdc6c9abd475a551b8ec212ba201303c02de65",
+        "token_ids_sha256": "6440d6812569b88a93b4fb5877d6a13138a10c627b39cd54adee08ddefe19eec"
+      },
+      {
+        "case": "arithmetic-02",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "3956026c853805d5dd053f1cb259f44626617e1c4946148db45c2561fea505ec",
+        "token_ids_sha256": "248a53a76b3082f4ca13c376452d9fa450cc372b27ee74681776106bf62f75dc"
+      },
+      {
+        "case": "arithmetic-02",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "b186fa6f03caeb78d03459674e39e7a017d68a025cb578ad0b68ebd10d082f31",
+        "token_ids_sha256": "d436f86318f7c2911a841f3dd3310ee76feffd7eecb427f87addbc87eb8c2607"
+      },
+      {
+        "case": "arithmetic-02",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "1580424514af0e654467483640add4ab4c970be8278db0a7dc51a6634160fafe",
+        "token_ids_sha256": "bc696d41dda4cfadb0539d3e6b16bba3012ff8adfd9eeba7f192b3fb020bd1ba"
+      },
+      {
+        "case": "arithmetic-03",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "758590c0e51a33e06f51fbffc5bbe0e3f777b0f7bf88ad708b58ebd6faccb390",
+        "token_ids_sha256": "bebe998965694a6a237aaa57377f49b2edab3815b787bd14a880e37433951748"
+      },
+      {
+        "case": "arithmetic-03",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ee6e110a809b16af00e88c49f620155ce1c738fd11950af79743146491eb0200",
+        "token_ids_sha256": "ece7b34196ccd3cee4f4533466992cc6819660af56bd1bb4c1e4a2efbd09d1aa"
+      },
+      {
+        "case": "arithmetic-03",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "69fdc96534e7e6c2b009822cfbe68f1b164bd8cc28a38396ac6532c9e3717627",
+        "token_ids_sha256": "290d3d9286cc2fa631a68f8d7d7c21e79b101dd6f9d47d2e5fed68cae8dfa9f1"
+      },
+      {
+        "case": "arithmetic-03",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4fc7455c07e88c95e3e4b9789d219e4bda857bbd8e75099da123855c41c93a9b",
+        "token_ids_sha256": "15384731525c414b4dd4496505a9d1c10db7b09e191ecc1ca1324ce51740b9c8"
+      },
+      {
+        "case": "arithmetic-03",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "758590c0e51a33e06f51fbffc5bbe0e3f777b0f7bf88ad708b58ebd6faccb390",
+        "token_ids_sha256": "bebe998965694a6a237aaa57377f49b2edab3815b787bd14a880e37433951748"
+      },
+      {
+        "case": "arithmetic-04",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c91b90926ba37c04a16a7e51163ed88366f6ad1f91eb7a4c40a8213b986f6f2d",
+        "token_ids_sha256": "e1706e27836de214be0261eff95fe8145f2945effdc517a7e9115ded3148a67a"
+      },
+      {
+        "case": "arithmetic-04",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "bce2488b1645bb08cb2a157697c4dbc6d634da4e9b1949a9f77e3e987d4ac3e4",
+        "token_ids_sha256": "52455a3c634132dfcc0673da9ab990c3def7dac61376fdc9614160b1172b10bc"
+      },
+      {
+        "case": "arithmetic-04",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9823550d5e522b6cf480c9d82a1726e3b6530437570093e24bf7a82920827c1b",
+        "token_ids_sha256": "1aa3c00b507cfb0fda234a8759b649edad21c2e167c4fa77e4632cad6837ed9e"
+      },
+      {
+        "case": "arithmetic-04",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "60d7fc3ef5d976266dfbe9fa89c6f82bd990fbf2c751aff527871070001fa549",
+        "token_ids_sha256": "9115c65f6cc99cbcf2ceccf86f3a62ef18d39afaa429b74cc9297894a6ffe1cb"
+      },
+      {
+        "case": "arithmetic-04",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c91b90926ba37c04a16a7e51163ed88366f6ad1f91eb7a4c40a8213b986f6f2d",
+        "token_ids_sha256": "e1706e27836de214be0261eff95fe8145f2945effdc517a7e9115ded3148a67a"
+      },
+      {
+        "case": "arithmetic-05",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ff43ba2decea43f8543fdd36dd6c5c536c8bb97a54e5890131890fbf0c66c869",
+        "token_ids_sha256": "c93f056f94513cd2779bf1ce8f298d406ef3b027cb9f3a7db4a09dd7592ba85b"
+      },
+      {
+        "case": "arithmetic-05",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "3c2c2497661577a7357fa67b079db6b11ff395265a222b0a32b9d1b8bef02b6c",
+        "token_ids_sha256": "82acbe5edfb07831dc3d554b178ded8e7bf7f53ff8e743c4a1907b4447f6894f"
+      },
+      {
+        "case": "arithmetic-05",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "81a352decd47f903266075bf8d7d21980e0d39b7f19befdc4ef2565d9ab27760",
+        "token_ids_sha256": "3abdcb06a624fcc9f3dc6593e04e92ada7b0666f675fac2a07fabddeebc2d7b4"
+      },
+      {
+        "case": "arithmetic-05",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0a9271c8cb020e0a54ed4cc1c04eb582c81d4cafa9eae887b06fcc44d4b44653",
+        "token_ids_sha256": "37341803c1c617340a309634f0e56d48636b27377a3b4c254086439a753500a6"
+      },
+      {
+        "case": "arithmetic-05",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ff43ba2decea43f8543fdd36dd6c5c536c8bb97a54e5890131890fbf0c66c869",
+        "token_ids_sha256": "c93f056f94513cd2779bf1ce8f298d406ef3b027cb9f3a7db4a09dd7592ba85b"
+      },
+      {
+        "case": "arithmetic-06",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d9d7029346d3fc3e6f37173c76ec3db19e8668a82698b1345e71b56ec802ce28",
+        "token_ids_sha256": "3b45ede402669923581a4b9464a66bded91059bf64aa29c0006bdd81990e01f8"
+      },
+      {
+        "case": "arithmetic-06",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "fe48fab3c6ca13c7db6caa901b0961f231160d5ebac912fe33187ac662b0d3cb",
+        "token_ids_sha256": "8f92b0168bc1370747e0e92234e3e472b1d84ee4ac7936c9c58faae6e9f48e50"
+      },
+      {
+        "case": "arithmetic-06",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "5d8e4b8ae6957fa3c0fd3c6243db43414cd8117cb3498cb40d077c1283bfeaaf",
+        "token_ids_sha256": "bd4d6941e778d63fe01e72f236a70078db1d5438d9d6b0dea085cf5a2d78b7ca"
+      },
+      {
+        "case": "arithmetic-06",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "50e45800a302d26ba7ae81e3301201636685303bd7161f4713d8620fc847d504",
+        "token_ids_sha256": "a3798779d6db6bb5aeeaa5faa0060644926b9e489488ae7e9318c23351717b50"
+      },
+      {
+        "case": "arithmetic-06",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d9d7029346d3fc3e6f37173c76ec3db19e8668a82698b1345e71b56ec802ce28",
+        "token_ids_sha256": "3b45ede402669923581a4b9464a66bded91059bf64aa29c0006bdd81990e01f8"
+      },
+      {
+        "case": "arithmetic-07",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d4abab4ea9a54502aacc0cc037e6a3d2392eb4f00620de2842db7fcabbb5c6ff",
+        "token_ids_sha256": "d4555520da558d82c23f683d6d8bfc85b3662713180da8e247f4db6c6b9a94c1"
+      },
+      {
+        "case": "arithmetic-07",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4634f6ef7ab3216baa6ae0839b583c8c493bddf057c97228292e1c6493febbe4",
+        "token_ids_sha256": "b20040ab0deeea513aeee03ff9e7eadc1af6dfb0c53cc32100297911de879bc7"
+      },
+      {
+        "case": "arithmetic-07",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9da13b3584b88529d3d4d3afdf40f9dedc2e3259a3575ff617c6fb3eb0279cca",
+        "token_ids_sha256": "666ba3da90ecdd0822ff3f0bd76faf857abd0464d497dde69be343644b84f6c9"
+      },
+      {
+        "case": "arithmetic-07",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4e3d704c3f85473c586455d9a99d16f244bb2dd8337b78319296b0a36ebe479f",
+        "token_ids_sha256": "09095dae80f1f9451b2a58c2aaba62d236e5c6edae31f9d592eea9f3a177f2a1"
+      },
+      {
+        "case": "arithmetic-07",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d4abab4ea9a54502aacc0cc037e6a3d2392eb4f00620de2842db7fcabbb5c6ff",
+        "token_ids_sha256": "d4555520da558d82c23f683d6d8bfc85b3662713180da8e247f4db6c6b9a94c1"
+      },
+      {
+        "case": "arithmetic-08",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "f7d3036cc3bc0141cca89e9879ad1a822638eafdd067d0502bf748d895fdbe71",
+        "token_ids_sha256": "e980911b51a3c2dc5739a7c33f3f003e083ebbfabf6f60e75791c7825f1e2a95"
+      },
+      {
+        "case": "arithmetic-08",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "84ee45d869045b2f96a220c89040a27c4340529b2f10ba5a362125d9b99255ca",
+        "token_ids_sha256": "629913233bfa1e8693053fb398886818deb1c269b49a30ff12d752326dfea49f"
+      },
+      {
+        "case": "arithmetic-08",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "10b71fafeee6fb4c5250eaca77c8d848e0377898e8c4e31e3610fcb31a4bbf95",
+        "token_ids_sha256": "c776b00461042b3c38e18d8dff79d402a4ecced8c64c67d622058041ef04cbc6"
+      },
+      {
+        "case": "arithmetic-08",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "537f1753a07185b06875abe3a1ebb3ea40cd514794ce71f4a6a083f2e0d11fb3",
+        "token_ids_sha256": "a5f82822b6d6af6c83758f5e173128bca3c3d50d290cacb42309156672ab606c"
+      },
+      {
+        "case": "arithmetic-08",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "f7d3036cc3bc0141cca89e9879ad1a822638eafdd067d0502bf748d895fdbe71",
+        "token_ids_sha256": "e980911b51a3c2dc5739a7c33f3f003e083ebbfabf6f60e75791c7825f1e2a95"
+      },
+      {
+        "case": "arithmetic-09",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "bc7d34da11dd014cb41eac5b8ed8758a7a88dcafccb274866a6812b62fef30ec",
+        "token_ids_sha256": "62c6cfebfddf285cd89ff58d1833e6e535fb72484b68251bb0d92fe37aa593e3"
+      },
+      {
+        "case": "arithmetic-09",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0bf19f315da38c20699e3eaebf5f02d9ed76193aa99eaf0bc9167dc210416d12",
+        "token_ids_sha256": "0bf26e146c4373d1613d6c9ec028b465601dd2173c8c874cd48e405c35abe5e9"
+      },
+      {
+        "case": "arithmetic-09",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "2a4aed936eaa1a57af5f4e11d45f1c93becbceed7bebf4349e2f5c21965cb183",
+        "token_ids_sha256": "cc54e72fe17312e9865a641818fdfc7e78ebe03daf8c53c2096f650d5cda29ad"
+      },
+      {
+        "case": "arithmetic-09",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "439ed4db5d4c179b66f0d507be08cd06fff0c5017d15f945c53127da81cab47f",
+        "token_ids_sha256": "40c23ec4db2ea5e8b6ed8a8591cdd74339f986c05f5474bcd4f928c8ba95fc74"
+      },
+      {
+        "case": "arithmetic-09",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "bc7d34da11dd014cb41eac5b8ed8758a7a88dcafccb274866a6812b62fef30ec",
+        "token_ids_sha256": "62c6cfebfddf285cd89ff58d1833e6e535fb72484b68251bb0d92fe37aa593e3"
+      },
+      {
+        "case": "arithmetic-10",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0b724f38e7bd9051ec7c968b1663b1b3635ec23c3b94c91e10aa9222fa7eadd9",
+        "token_ids_sha256": "31ef5951966d506a875595669d28381adb7f3eb8ca36e728f849aa0d8458b0cd"
+      },
+      {
+        "case": "arithmetic-10",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "6062a96149481c62ced98da104245bb90e591cb34292c12eadfc3d9fb67299a2",
+        "token_ids_sha256": "80b76dc69c7844e058fe435e5c28373408aaf2b9f5dc24f6aafc735cf75b4b61"
+      },
+      {
+        "case": "arithmetic-10",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "884f8cf268a54b70f40569b1a039037963471f5ece48c681f117920ccae1fdea",
+        "token_ids_sha256": "b6fabf136f2e5a33c007f475cbd3b7f87bdce0cc1da9196ea00b679d62ce151c"
+      },
+      {
+        "case": "arithmetic-10",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0dc637f9d7ed2ed1fca6922a413f5c47ebf3ab6579aa31b4ce3bff07af300ab9",
+        "token_ids_sha256": "4363a2a13106c422c12752f4b87b3581eef342b6a2340ada93f635851c45f37d"
+      },
+      {
+        "case": "arithmetic-10",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0b724f38e7bd9051ec7c968b1663b1b3635ec23c3b94c91e10aa9222fa7eadd9",
+        "token_ids_sha256": "31ef5951966d506a875595669d28381adb7f3eb8ca36e728f849aa0d8458b0cd"
+      },
+      {
+        "case": "json-01",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4de0442df7ac09fc61d43b9ffa738f4f2d3573f71ac77dcec22aba786bc0342e",
+        "token_ids_sha256": "e4b144f0c08aa87b573f59f54bf96d10641d30ddfb373ad355bb3938075682c0"
+      },
+      {
+        "case": "json-01",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "5f50a83fafc47f078151ce396bbc1ba2fb64795e38165ab27adb8513a9aae264",
+        "token_ids_sha256": "bbbfd4ba0e5b27f3914e50c9be5a2bbf8c7cbedc83bd8ee3ffccf36090b042e6"
+      },
+      {
+        "case": "json-01",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "31a143c5a416e7f3a93ba3739fd66634ca3a728482405636c7048ec40161619d",
+        "token_ids_sha256": "13a83c0266c974090ef5736aa8cade8685638d64637e2787534440dab7152c6e"
+      },
+      {
+        "case": "json-01",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9dcf381dbeb544ab3054db3afdc5ff7d33833027519c5c03d98d59de2d76b9c8",
+        "token_ids_sha256": "7debf297e3e6f800a92cccece7deecb3f1edcdd30c80ca6beaf3194c66c267da"
+      },
+      {
+        "case": "json-01",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4de0442df7ac09fc61d43b9ffa738f4f2d3573f71ac77dcec22aba786bc0342e",
+        "token_ids_sha256": "e4b144f0c08aa87b573f59f54bf96d10641d30ddfb373ad355bb3938075682c0"
+      },
+      {
+        "case": "json-02",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "f52a34ad0f6b545d6d81c4ccd7ec54604a733c172a8f4f13e51fa28ef276d803",
+        "token_ids_sha256": "1eb009c158c67adfb279a1d55cfc67b2559b54a9ee95b6771952fde76795ad6b"
+      },
+      {
+        "case": "json-02",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "1a2f1c8bb1b3286ab8a957f85f575a0d4d5a0f1bb971b742e69c4990fec9577f",
+        "token_ids_sha256": "09f7a63ef1188c209d80f1a7cc6ccf071e931b268cd115601c98ae083609b5b9"
+      },
+      {
+        "case": "json-02",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0fa5d71515f9c8233ad41e135dc72cf6cca670d4a56253727155ec1071ce271b",
+        "token_ids_sha256": "a630c470d9a5e52e1e429dd2b3c84cadbfbbf02803317d58008c99d1fe4c51e2"
+      },
+      {
+        "case": "json-02",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ea329dcbf37dc55489fd11c37d464c2bc3ac63e0e9a81b8c56d3d875fcec3a74",
+        "token_ids_sha256": "e940fea90dfa2c71564f797a50cb8dad160f993fe006bc1a4b25b1f27f603a9c"
+      },
+      {
+        "case": "json-02",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "f52a34ad0f6b545d6d81c4ccd7ec54604a733c172a8f4f13e51fa28ef276d803",
+        "token_ids_sha256": "1eb009c158c67adfb279a1d55cfc67b2559b54a9ee95b6771952fde76795ad6b"
+      },
+      {
+        "case": "json-03",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8a3dbb38d2827e057168fb8a3d6a2de8e032ec6852a8e87c0b361543f8f6056f",
+        "token_ids_sha256": "b34856c2e771c03911bab8353efb901ca50653e99bb8fa22c52970ca9437387f"
+      },
+      {
+        "case": "json-03",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "37bd9133255a681630a15fbf5f902a6c60689d4b6cb7a7fe1631eac26bda9142",
+        "token_ids_sha256": "6ca400f3330a08f46318460cafd0f658af05cb80994f132a306a291dcc6e3aea"
+      },
+      {
+        "case": "json-03",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "15eda6413cbfe554fd979fe671f9b29d3460190424f4fad532170efd3eed6fee",
+        "token_ids_sha256": "d8bb4ed4c951d432251f174077cf19e53716afd602d49e5fa3a561cd9c7e0324"
+      },
+      {
+        "case": "json-03",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "e13d3dcfb7025ae479066fa806518c2e558c8f3c154d0c907c6263e67349ca95",
+        "token_ids_sha256": "1e5cfb4792450de79b24f3850daec9b92b2aaeaf090e31c9c2d0ec65db03f1a1"
+      },
+      {
+        "case": "json-03",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8a3dbb38d2827e057168fb8a3d6a2de8e032ec6852a8e87c0b361543f8f6056f",
+        "token_ids_sha256": "b34856c2e771c03911bab8353efb901ca50653e99bb8fa22c52970ca9437387f"
+      },
+      {
+        "case": "json-04",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "1f303eab286c8e0b2dfd1e037488ced9ec2ad631c83baa367f2e7ce84c0d754b",
+        "token_ids_sha256": "7ba01d3705a9a2fc2c44adebb589fe7c4cc55d32f4f34b4c171b0e004c68cfeb"
+      },
+      {
+        "case": "json-04",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8653dbe405bebf2ae11a4e03ce412656fe2c8edac19590eddd7683ff0a161a41",
+        "token_ids_sha256": "12c6581c2d8aca475d83a32e06a2c4c13acaebe759c2559bb9d0b0072bd01eea"
+      },
+      {
+        "case": "json-04",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "390532466ee5550aeeab6067168146cfc47eaffb3423ff6758df6d6733a4d59a",
+        "token_ids_sha256": "9ae37ad8259608f3968e7aecd6ae80cca934d185231a2b18ab6dfd0ff50dab36"
+      },
+      {
+        "case": "json-04",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "33f30604fafeb5471321edce74a26ec52381f03eabfd6acfa4a79784124d9855",
+        "token_ids_sha256": "9fb15ad820d73935b93d25cc15c9277639804a93d7fca097cfaa1704ff4862e0"
+      },
+      {
+        "case": "json-04",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "1f303eab286c8e0b2dfd1e037488ced9ec2ad631c83baa367f2e7ce84c0d754b",
+        "token_ids_sha256": "7ba01d3705a9a2fc2c44adebb589fe7c4cc55d32f4f34b4c171b0e004c68cfeb"
+      },
+      {
+        "case": "json-05",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8ed29d2b59a460fa3e6dbee839a9635edbbaf50eea80b88bff5eacdc78e73fc1",
+        "token_ids_sha256": "4139c153e92d79a41aea986199662c5f31289a0f8b774855c7c4f66dc6b8db17"
+      },
+      {
+        "case": "json-05",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "a3818397f7583a07006ca0feefadf46f8f09f4f14123cb7e61e2eed7b3549b5e",
+        "token_ids_sha256": "8184a5cc8b187c66285b51417a2655460c534ae8a98581f6abbeea0db3d9d7fc"
+      },
+      {
+        "case": "json-05",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "54972f2f1eef05bb9bb6905aac8573dc4e6074740cf6e74633295c307d00e256",
+        "token_ids_sha256": "331548692bf3d1c450ffe65bafac49565037ecb0bc078b03b794d88e3a55395f"
+      },
+      {
+        "case": "json-05",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9cb233dad0e7c052b679c8abcbd1c37f806f8771ddef95a255739e1aadbbeb30",
+        "token_ids_sha256": "d9a08463d3373938a3b098b85182894715e811c42f08dea12d3b3c699d65696d"
+      },
+      {
+        "case": "json-05",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8ed29d2b59a460fa3e6dbee839a9635edbbaf50eea80b88bff5eacdc78e73fc1",
+        "token_ids_sha256": "4139c153e92d79a41aea986199662c5f31289a0f8b774855c7c4f66dc6b8db17"
+      },
+      {
+        "case": "unicode-01",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "873831a91e22b06c76fc5f676cf413eb134887725518f0915bb5e84d7f06b2b5",
+        "token_ids_sha256": "f3baf0dc6982eaece8c4cdd4c4671b1ad756619ae32530bd08481c691a4986c8"
+      },
+      {
+        "case": "unicode-01",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "e73da63a506d2412fe58c20add9e44337afcc80e82ea3f8149529b0a5be5c989",
+        "token_ids_sha256": "b556477b09485ca4d3c36829cd8f61f1145fae0b1eb1efef0aeab02853ceae18"
+      },
+      {
+        "case": "unicode-01",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "749b9ab5744f103026bc7b97cb021f820aec8a708ef3117ed9ffb3c82dc727a6",
+        "token_ids_sha256": "c04c84f9c01ef219732a2a5a12df123c7596e297e5c286aa0929648a7e3f4a71"
+      },
+      {
+        "case": "unicode-01",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7222991af7ce9ca6c332f1004b1447347009c19ad2c6e305872c9b885e58ee6e",
+        "token_ids_sha256": "c4e35e4934451ebe9cdff2d460d661d4c2f4d2b10cfc3885a233d1aca0671af3"
+      },
+      {
+        "case": "unicode-01",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "873831a91e22b06c76fc5f676cf413eb134887725518f0915bb5e84d7f06b2b5",
+        "token_ids_sha256": "f3baf0dc6982eaece8c4cdd4c4671b1ad756619ae32530bd08481c691a4986c8"
+      },
+      {
+        "case": "unicode-02",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7d29c826bee76e345dcad80d469e0c238f48af54819f1912de0f6038169538c7",
+        "token_ids_sha256": "4a3f23fdbd290986351a2d30596196a47868493013f89a206e7a94d6ada39ad4"
+      },
+      {
+        "case": "unicode-02",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "2941f77f6d53b2f4123fedbb77ecfc009e7da0970021af7cbdbf5ff5d5e27749",
+        "token_ids_sha256": "08c986d3979914a9a4ccc7205f9acb2969db0063df8ed2eddf8850891b3b2ae8"
+      },
+      {
+        "case": "unicode-02",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "abf48e9b182940d04cd3ae1b89bbe6d2558044551682fc80660e82b6cf7270d8",
+        "token_ids_sha256": "662cc9514c92de9469ecb2c47fedfcf0f1468d481445cbb5fa73950e6ca8310d"
+      },
+      {
+        "case": "unicode-02",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "998ab000c7531bd04a91be93ebbae3b10bf0128784a4dcbfad895249f02de0dc",
+        "token_ids_sha256": "306a80c5aeb1708c537527547590859a3adcfa5951fb9937ffd177b0349a5199"
+      },
+      {
+        "case": "unicode-02",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7d29c826bee76e345dcad80d469e0c238f48af54819f1912de0f6038169538c7",
+        "token_ids_sha256": "4a3f23fdbd290986351a2d30596196a47868493013f89a206e7a94d6ada39ad4"
+      },
+      {
+        "case": "unicode-03",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "bf4b5f71c04f51824c51cc1426cfcd0c437171656133c7c369c23c357223b486",
+        "token_ids_sha256": "e00642a1e34a12ef656d09970b2324751d6af2271b01c433d7d1b16e56740d9f"
+      },
+      {
+        "case": "unicode-03",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "12fbc9a75e8f80f90e604940aec261dce857bd6a9a29b44155c780b7da72da34",
+        "token_ids_sha256": "a2974c301f5ce702b9c5c3046d5f83c4c435d41bed622d4c48874b40a580861f"
+      },
+      {
+        "case": "unicode-03",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "6add51b230cd93bf48a49f4db9373b3ec1e551d2ec96a3b4dbea5a7d015b144f",
+        "token_ids_sha256": "8416686c5324dacd87e006bae5782e6bbf64b80d63173dab546e2e9818d3f641"
+      },
+      {
+        "case": "unicode-03",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "fb1f80d916cbd5dfb8280688956eadd738271c2bc2337b80815bb9fc51731c32",
+        "token_ids_sha256": "b81d42ed82db2354b3340df6024f1231000a5325ccbb914db8358283a0532e62"
+      },
+      {
+        "case": "unicode-03",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "bf4b5f71c04f51824c51cc1426cfcd0c437171656133c7c369c23c357223b486",
+        "token_ids_sha256": "e00642a1e34a12ef656d09970b2324751d6af2271b01c433d7d1b16e56740d9f"
+      },
+      {
+        "case": "unicode-04",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d7142c4d4ade7ba1b87072e8aacf73c17f08e255dea6eb1b06a621353211c5a1",
+        "token_ids_sha256": "4ed5ff2c7b6a2f8ae5c264e871a1d758a316b2f28f5870331b4458bfe1d8caf0"
+      },
+      {
+        "case": "unicode-04",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "cc557c74d6c9a5c41b9ec54c4924315e2976080858c08f18c03110223f8aead6",
+        "token_ids_sha256": "f0309c2a3ee2ca5a83c54e5b1c23482a075686d5e997824788a572193bce952f"
+      },
+      {
+        "case": "unicode-04",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0f4e07bf9c8129f5ecfe3c815adff8d78d3013cf8157d85f113f6035966762cb",
+        "token_ids_sha256": "f0c31548432c48ba755043292d19a9dcc64e81677e3a0c4129b663d664df1fe8"
+      },
+      {
+        "case": "unicode-04",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "515d47b9366bdbff3c4da92f763fcdd382487ae4dbc97c88dc4f677e600ffdd7",
+        "token_ids_sha256": "14f6bce785b2517a53fc15d776f89d688fac43e4b11d434a63903fb451628e3d"
+      },
+      {
+        "case": "unicode-04",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d7142c4d4ade7ba1b87072e8aacf73c17f08e255dea6eb1b06a621353211c5a1",
+        "token_ids_sha256": "4ed5ff2c7b6a2f8ae5c264e871a1d758a316b2f28f5870331b4458bfe1d8caf0"
+      },
+      {
+        "case": "unicode-05",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c29c6cf2be9160e98648fa8dc071fece6ea6660cf69f08e1f7b0aa54fe316b7e",
+        "token_ids_sha256": "25632625ac135dc1d449c78798e64ada84f0ae48804aafbc09914e8b026d3146"
+      },
+      {
+        "case": "unicode-05",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "44b83f63cdda91288fca02c6155eea41a474b6cb2fcdad68ff9587cc055a2fab",
+        "token_ids_sha256": "1fd9dc706e9b32b1af44b066e7abe40afa5f6636a645a3c9879f9dd0d6cccf6d"
+      },
+      {
+        "case": "unicode-05",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "54513566862c2e37446c1bf5d746b65b4c32e938f68764673ee100ab40c0f71d",
+        "token_ids_sha256": "c7871f918bec46deeddf3630e2fe8a1b39970eb5b8efb50c135b2ea9801c82c3"
+      },
+      {
+        "case": "unicode-05",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "0857bea7133e43b4ef19f718421ae2a03622f96d921ab05042503fd15895d2dc",
+        "token_ids_sha256": "ebf3da501675389d42d06b41329a53a3f90650cc2ec38c8319c9ae9088ffc866"
+      },
+      {
+        "case": "unicode-05",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c29c6cf2be9160e98648fa8dc071fece6ea6660cf69f08e1f7b0aa54fe316b7e",
+        "token_ids_sha256": "25632625ac135dc1d449c78798e64ada84f0ae48804aafbc09914e8b026d3146"
+      },
+      {
+        "case": "roles-01",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "69197243135d48e4810d294922e7cf658a4be3dee7774ef4c990645dda560c23",
+        "token_ids_sha256": "678d6c1d5b98f94c37677b7eccaf46dd0cad48028bf1da9d26ca95b8bbb5c07c"
+      },
+      {
+        "case": "roles-01",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "be4e59de4ee724c6ad88fa7bd5ba7e3b720d10007ddb784bf816c159a5177638",
+        "token_ids_sha256": "dedb5f6a296d441e6ae9f0800da3ff20185aa7679704d810bee4f7e4cb688866"
+      },
+      {
+        "case": "roles-01",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "b18660722791b356c87ecaca8b311901c0c1c7998c333b9f1fb4a30e4db71cdb",
+        "token_ids_sha256": "d814d754001623d861b248019f7b6217520c75b930e0e35bba3c17e58cb7d5fb"
+      },
+      {
+        "case": "roles-01",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "eee7dec1e6469d438b5d21989ef21095f287efd1eae13b4734fb56cfb84f1959",
+        "token_ids_sha256": "ceeb56c216d9df66e9942193cc03ad169bfa40959d8d381188443e2006f8ebda"
+      },
+      {
+        "case": "roles-01",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "69197243135d48e4810d294922e7cf658a4be3dee7774ef4c990645dda560c23",
+        "token_ids_sha256": "678d6c1d5b98f94c37677b7eccaf46dd0cad48028bf1da9d26ca95b8bbb5c07c"
+      },
+      {
+        "case": "roles-02",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "712d00c38b9e83916c4835a7fd39282d23ac9320c9f19bc16c16eb4a4f040011",
+        "token_ids_sha256": "e2bf0f7a3aae6e78f914021f4402fea5d91d6c707ce5a4cbf620829cf3c0764e"
+      },
+      {
+        "case": "roles-02",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7fa28a820c16c0cdb7bd520ce67584863bc2621963b5ca86b0ff9b31bb4e595c",
+        "token_ids_sha256": "2cd4c83c774cd9f5d4cd6cb831fc524e56177ada3fe18441cb5dc90c9ed40a91"
+      },
+      {
+        "case": "roles-02",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "13e1946a670a1e3fb4f611341760b334e9a0e451ae65eccf52b2a6d03e4fbe61",
+        "token_ids_sha256": "6ee01320aa8b8230d8d9fce35bdf2cf5379524e389f2fcaef27b4a8c1d8360bd"
+      },
+      {
+        "case": "roles-02",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "040d17c8416b169af8829d4ac474fb2f59b4df2c3a1e05b97798f97b702f5d85",
+        "token_ids_sha256": "bfad556e9b5da0ccd563f009ee1b2b332165bf7e740db9118d0b3293d34227b9"
+      },
+      {
+        "case": "roles-02",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "712d00c38b9e83916c4835a7fd39282d23ac9320c9f19bc16c16eb4a4f040011",
+        "token_ids_sha256": "e2bf0f7a3aae6e78f914021f4402fea5d91d6c707ce5a4cbf620829cf3c0764e"
+      },
+      {
+        "case": "roles-03",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "3b85188652babd19a119f9bb6970d704077a93af91a2b0d16f24a787188a4068",
+        "token_ids_sha256": "d94af0af1c092b4782cba4ad4b25b38414db15ee002a375a2fd0c75ba6d05dd8"
+      },
+      {
+        "case": "roles-03",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "ce4853f4464de61b59ff75e37680c4dd1ae59aedddebcbea809f3a2c4f1affca",
+        "token_ids_sha256": "12248951e0fdecd25101bc14b252df89f96b821450e55e896c8f6e12615206a8"
+      },
+      {
+        "case": "roles-03",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "dfb78b9b5145697a1185231bc1af8f91785880221c5da5154fb7f91459ebd927",
+        "token_ids_sha256": "a61aacd7a56e3427d13f7c06794de7b3903885140d063a8a9373ed2d21507c3f"
+      },
+      {
+        "case": "roles-03",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "d9381db20bf9a23be308fc95f4ff07b7af74b610c2c2c3daeb886c545bc647fc",
+        "token_ids_sha256": "42408fa09b988dd65ee60fe0f098d82323f1a9a3c60e2067263549573d9e8a8a"
+      },
+      {
+        "case": "roles-03",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "3b85188652babd19a119f9bb6970d704077a93af91a2b0d16f24a787188a4068",
+        "token_ids_sha256": "d94af0af1c092b4782cba4ad4b25b38414db15ee002a375a2fd0c75ba6d05dd8"
+      },
+      {
+        "case": "roles-04",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9cebda0d117eae0f377f38014a909a8ff59a7c9813803ab4be09faae59f3edff",
+        "token_ids_sha256": "00ab1f28e66ffa06478fa3d42dbe05c1b7b243e73cc1324147361bc2e71fcf27"
+      },
+      {
+        "case": "roles-04",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "78fb14a7cf8badcc8b7c4b2dc1deb5f18d49d1c2e08f682b3508982f9274d5f1",
+        "token_ids_sha256": "395aca8a9be097ba2cdd5e11bcf9fb67928fb9ec72fc39e9097414249f120f56"
+      },
+      {
+        "case": "roles-04",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "580a0ed251e3ae771fe6d9dc098cc46198d1512615318b3c360c0ad93c7f0b37",
+        "token_ids_sha256": "d03a56676074998e963716039d4fbeee0404bff282283e3eba2bcdc534e48527"
+      },
+      {
+        "case": "roles-04",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "43603f71f3d781afd5495449e11ddbcfb123e4bbfcb91d49127b8f3d38c1d525",
+        "token_ids_sha256": "e0ef1daabe4baf72ddcaf1569e2896647a3170e7eeabb5bfcf56df297a19f300"
+      },
+      {
+        "case": "roles-04",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9cebda0d117eae0f377f38014a909a8ff59a7c9813803ab4be09faae59f3edff",
+        "token_ids_sha256": "00ab1f28e66ffa06478fa3d42dbe05c1b7b243e73cc1324147361bc2e71fcf27"
+      },
+      {
+        "case": "roles-05",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "221adbad330129783e9a8df3a43fa0f498b8bcbfb3812c655de79639ec91790b",
+        "token_ids_sha256": "d7a18d6d1252cb2703786b18e1d3eaa16a9648eea36b93a1ded2006f85a37091"
+      },
+      {
+        "case": "roles-05",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "12516193a253ae877876f64b85f7e4b494e26eea096266af085340cb58053c64",
+        "token_ids_sha256": "e2cccb2f8693bc68d97b17c93bd70a633deaa59bce1000d60277300e8bec630a"
+      },
+      {
+        "case": "roles-05",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "accba5632f386b18f07a489f2533b56bf288f4dacc67092747c164c02c8d1410",
+        "token_ids_sha256": "320f0dae8da2a7da89b861a88cbdb0681c059aa2c32becbfd1ca61cd1d51118c"
+      },
+      {
+        "case": "roles-05",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "cb560837a0396a76b67755d82e2f085f3579679fee20d2439cd3091e12c021f3",
+        "token_ids_sha256": "ef56d5a979147ce65ac1fdd78283b8f25c0a80d8321e1d41b92c512712cd7506"
+      },
+      {
+        "case": "roles-05",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "221adbad330129783e9a8df3a43fa0f498b8bcbfb3812c655de79639ec91790b",
+        "token_ids_sha256": "d7a18d6d1252cb2703786b18e1d3eaa16a9648eea36b93a1ded2006f85a37091"
+      },
+      {
+        "case": "coding-01",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "051b7aa8df5fde373f9f96fca41c7d670d7b11fcd2ddb142a390bea76999b5f8",
+        "token_ids_sha256": "b1ad5965c18fb4f6e2b2fa341dae0f025c989df9bf0e3283189e551a7b96da82"
+      },
+      {
+        "case": "coding-01",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c6fc7d3512acc9d7d7bd3c0788be368e86ddf7cee0036b70e86e293a10e89596",
+        "token_ids_sha256": "d095663e2e962f0090c4e595ae1454d546a5528db6aeb791bc4c90ef35117d40"
+      },
+      {
+        "case": "coding-01",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "04300fbdce4d473786e1ad0f4fcc7e198d4bcdef5f266c1ad05fff21cd456200",
+        "token_ids_sha256": "c62e1d77b8b548ea85c60177e753441b267e5935ce5a017139594a0cf104d456"
+      },
+      {
+        "case": "coding-01",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "712e6076d40ff56ab6beb8b543193f01de8ff494109f916e79e351e9051c2bcd",
+        "token_ids_sha256": "0722a8a5fb8124f7a028af8507b238963d8bc82429a6726fed54c4694a8cce04"
+      },
+      {
+        "case": "coding-01",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "051b7aa8df5fde373f9f96fca41c7d670d7b11fcd2ddb142a390bea76999b5f8",
+        "token_ids_sha256": "b1ad5965c18fb4f6e2b2fa341dae0f025c989df9bf0e3283189e551a7b96da82"
+      },
+      {
+        "case": "coding-02",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "411d173959d148e172d44867aef97f6367f1208c5086c1c656b8c61bdaa1ad97",
+        "token_ids_sha256": "c21a74da2b8d69aa0a51e372ee4bb83480a380d34c8a5e5c8e9cbd437ffd456c"
+      },
+      {
+        "case": "coding-02",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "580edec3aba6a11165ebf4fdf0740108cc3d7b0382cd1e4aeb46b4ba749cfc8a",
+        "token_ids_sha256": "e6a92c6cf675c02b37b416cfb3e7dffa09e52a2260cbc79165a2073f6a09ec13"
+      },
+      {
+        "case": "coding-02",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "7e86786de23289534a245e624822a9e0de18e1e9fafdb381cd7a833d840b474b",
+        "token_ids_sha256": "f69df4a17b6dac41f1b470bd58a9b05b5c396b5b763f0d658b26b5fd9ab2f2f9"
+      },
+      {
+        "case": "coding-02",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "bbcc8f29071b25d8fb0b525aa9fcdbfa344f77559069b617a5883ceefac21dc9",
+        "token_ids_sha256": "29a24edf2946e8334424a9eb1c654efbd861c71def8c72e9cbd43cd3f25e282e"
+      },
+      {
+        "case": "coding-02",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "411d173959d148e172d44867aef97f6367f1208c5086c1c656b8c61bdaa1ad97",
+        "token_ids_sha256": "c21a74da2b8d69aa0a51e372ee4bb83480a380d34c8a5e5c8e9cbd437ffd456c"
+      },
+      {
+        "case": "coding-03",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c8b31bcbabdc40e027ff96975825c9e8e1c707b34ad353efa4349d1eeb105e43",
+        "token_ids_sha256": "a0c8029475104f3da2c240b437c2611571e30faa137278d9dbb144ae8a69d719"
+      },
+      {
+        "case": "coding-03",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "6a2ced8824a9a8b8f5198c6a05c27504fa89fc64fbec2ecb6ad7dc60945c3f01",
+        "token_ids_sha256": "40e5461c319001a2e97fb9bbafd7cb867634caaecdc07b65c941415002402daf"
+      },
+      {
+        "case": "coding-03",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "3830b62bb1d8d8bb3ae2632340987db3c294e516bb169422217377831e816b4a",
+        "token_ids_sha256": "de27378bdd6efb851642482fa6dfe3f11d39c52265218c6fec331890b0945b55"
+      },
+      {
+        "case": "coding-03",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "dff5aa85158115d29350b91444084fac347e49f1f8286ecd3ffa1875e66a7594",
+        "token_ids_sha256": "3349f46bfc31fc3682bfc4c96ddaaa3738c886eb37c61e7464f5887f74a2efc1"
+      },
+      {
+        "case": "coding-03",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "c8b31bcbabdc40e027ff96975825c9e8e1c707b34ad353efa4349d1eeb105e43",
+        "token_ids_sha256": "a0c8029475104f3da2c240b437c2611571e30faa137278d9dbb144ae8a69d719"
+      },
+      {
+        "case": "coding-04",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "525527a25652337c7b127cc17ef22ff8ab41b8624bc5231067ae32d9f7a83f39",
+        "token_ids_sha256": "fc9272e512c52fb5862355801f76a9f0e1fa8612c04397f30b63b85bc0edd39d"
+      },
+      {
+        "case": "coding-04",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "072168df13573a942e2940693badc6bafa7f11669572d6fc2f229001b4469653",
+        "token_ids_sha256": "a66ea75e01bfc1615c9f537012258e3f087b634feb08fa4ffd8d8ba340e0f305"
+      },
+      {
+        "case": "coding-04",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "208f77c8f98b9ee4f04a355793762383877e67d6aaa289b2dfcf1bd75fbddebd",
+        "token_ids_sha256": "40c119c2ad05ffe426d8e69e5b341ca118babdcf8b7818b236d0d352d1d58741"
+      },
+      {
+        "case": "coding-04",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "6e7390f1ec457eda8db0918223dc25c095c8f4172ea6b09b831437185f02c427",
+        "token_ids_sha256": "0499243975895f7c92a2d18281d1f17e2c36c5846b221d289fe34306c39df9e2"
+      },
+      {
+        "case": "coding-04",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "525527a25652337c7b127cc17ef22ff8ab41b8624bc5231067ae32d9f7a83f39",
+        "token_ids_sha256": "fc9272e512c52fb5862355801f76a9f0e1fa8612c04397f30b63b85bc0edd39d"
+      },
+      {
+        "case": "coding-05",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9f2ad75fb0b84ce9efd65a3efb8010b701fcf92941cde2ba6a6ef20940bc8ccf",
+        "token_ids_sha256": "4bd80bdd05badf800ad6c69d84e685cb702dd393a4409cffded2e4579467560b"
+      },
+      {
+        "case": "coding-05",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "cd8d48627da73de9ab073e5d0622f81210cac59854b54da2bf2aae54bf55370b",
+        "token_ids_sha256": "eaae104ce5b9c9eb5a26f34ce473f160fd29f32830097ac8d142f51f2e5827cf"
+      },
+      {
+        "case": "coding-05",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4e769996021bd0ea7f66d57b712c999869e42e96463efa83e93369b6b60fc86e",
+        "token_ids_sha256": "a7586d4d012fb8954d08416f41f08b3cf130aa756497f3f20cebbb7ace39f2b1"
+      },
+      {
+        "case": "coding-05",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "50912990af564be675422742d3f8d089638d94c8816fec6c89cbe50e289247a8",
+        "token_ids_sha256": "33732ad475a97239de800bf23cff2107361231851a056386d34492ad51bdcbcd"
+      },
+      {
+        "case": "coding-05",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "9f2ad75fb0b84ce9efd65a3efb8010b701fcf92941cde2ba6a6ef20940bc8ccf",
+        "token_ids_sha256": "4bd80bdd05badf800ad6c69d84e685cb702dd393a4409cffded2e4579467560b"
+      },
+      {
+        "case": "coding-06",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "82b4aa32308ea9227904a64f15d1e6d04c19de98e6ac96c29e48c9b128ec06fd",
+        "token_ids_sha256": "267b2aa4000603afed98cede64f743b1d7595e2b821e8a3fcd3ff4e508ddaedd"
+      },
+      {
+        "case": "coding-06",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "81fcfdfd7d5fcb963098331741f221d85160d3282b0ec492a8b12e3bd21a574d",
+        "token_ids_sha256": "bfe93c4c4646449437207262ad6b630c0f47d7099de75dc72f39834928f9ea12"
+      },
+      {
+        "case": "coding-06",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "4d5147f44a60a4825fd8a217b46811224cf80183f9d53f5ce6c8cb9788aae7bc",
+        "token_ids_sha256": "0dd4cd614a3f9d968a52cfbff7d2db4d8e8efebc9d91d8bb48a6f3753c253ea6"
+      },
+      {
+        "case": "coding-06",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "06189fca5fb668321f952582429ceb9d88c42c9636b4b355d5d3a3e553f091be",
+        "token_ids_sha256": "f707a9bf1f7e54764c1d9d85b6249e4476477f0dd2740e0d79ce1c9743dc063c"
+      },
+      {
+        "case": "coding-06",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "82b4aa32308ea9227904a64f15d1e6d04c19de98e6ac96c29e48c9b128ec06fd",
+        "token_ids_sha256": "267b2aa4000603afed98cede64f743b1d7595e2b821e8a3fcd3ff4e508ddaedd"
+      },
+      {
+        "case": "coding-07",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "b0d7e700a9fac7b48229de675455ed2940caa11f4b1cc49c209a37301a6c2b8f",
+        "token_ids_sha256": "c71180eb108ca429cb4ca6dffcab936a64ab26f63badea18129a4ab6c940d1e5"
+      },
+      {
+        "case": "coding-07",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "e4bcb00c59182c545871cd2c67195403005b03152d8796bf0beae9c047dcd617",
+        "token_ids_sha256": "c56e5386106e5377b9e81ee81fa50c575d4088c26b9971cdd59f76fb6e85cb5a"
+      },
+      {
+        "case": "coding-07",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "5caae3bccbe931fe69c7b7fe5005227be953a288044615e8314a64c6955200a1",
+        "token_ids_sha256": "db7915bbe12598651b963d0cf689f970021965a932800c1b982695c305803c32"
+      },
+      {
+        "case": "coding-07",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8a6a197dceef434429c6436feced1f1e6d1d9bd8365c1b701164c25b7ad1019c",
+        "token_ids_sha256": "d18ff1d4baf16bb9fc113e3880da3d144bfe41318982afc841747d1e1fa210cf"
+      },
+      {
+        "case": "coding-07",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "b0d7e700a9fac7b48229de675455ed2940caa11f4b1cc49c209a37301a6c2b8f",
+        "token_ids_sha256": "c71180eb108ca429cb4ca6dffcab936a64ab26f63badea18129a4ab6c940d1e5"
+      },
+      {
+        "case": "coding-08",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "6450b7356a6c99f5fa6ee088f4cd053ea337b1a2acc88f8e02f9b3cd5d55e0dc",
+        "token_ids_sha256": "3c5750c4c0067dfc14491dcb8fcece2fd17c0b7803757705ffa9e148b5448df0"
+      },
+      {
+        "case": "coding-08",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "f8c20776c7229a2a1f91e51a268039b936f8146a12f2b6db1d8dc69d06655a88",
+        "token_ids_sha256": "567bd1194c16fb99264033311094691bb98e2682faf3cf9a3c2fd205b6f123c4"
+      },
+      {
+        "case": "coding-08",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "be22db21ec1ad4e3462089e8a95c744b8b9e0f759efebbaf1da7cecd4f47c2d6",
+        "token_ids_sha256": "de26d7e88adda3545ad0dbeb5bd563c07df82e0e33e2a16e3d7a510d8a64f36b"
+      },
+      {
+        "case": "coding-08",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "393e28f2aabec5fbd8eadbae4c2143d4bc31bba3810ade122b069dab5a3348a9",
+        "token_ids_sha256": "0660e04b0b6c1a679a7aa1ba2a8d25aed5bce7953a4e808628020fa9e78377f1"
+      },
+      {
+        "case": "coding-08",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "6450b7356a6c99f5fa6ee088f4cd053ea337b1a2acc88f8e02f9b3cd5d55e0dc",
+        "token_ids_sha256": "3c5750c4c0067dfc14491dcb8fcece2fd17c0b7803757705ffa9e148b5448df0"
+      },
+      {
+        "case": "coding-09",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8087587422b7ee2873c562eb9c4e06eac2a07978ccf0b937a29c02fc3b4c2356",
+        "token_ids_sha256": "bcfe76e47c160f14ab2d450be69d2323d7937287d3f5f9c1f534dc8a9d7efdb6"
+      },
+      {
+        "case": "coding-09",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "e18a2c027cdedd8cda071243f4be301285a09f18a3369d6ede946a304d930aae",
+        "token_ids_sha256": "dc1b5e6e60fdd50123ad7e05ec4cf7b82469e54b47bc00b123eb4f1de8cf1e82"
+      },
+      {
+        "case": "coding-09",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8d8bdcc536f5c82c277280b4a64e08d1d326a7c073a5255304f0c6de30da5e65",
+        "token_ids_sha256": "cd8a8a5bc78a2f006d9feea1d4a338d567a851dfe999e89326a65aac3dfe7486"
+      },
+      {
+        "case": "coding-09",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "897bc62dc9b88b47342347d094c357b25197a7686e3931356184f20228a0be89",
+        "token_ids_sha256": "449614cdc8ddbe692df31591b26e4cbb9f97ff69ebcdbe2785fcbc76c74a52ef"
+      },
+      {
+        "case": "coding-09",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "8087587422b7ee2873c562eb9c4e06eac2a07978ccf0b937a29c02fc3b4c2356",
+        "token_ids_sha256": "bcfe76e47c160f14ab2d450be69d2323d7937287d3f5f9c1f534dc8a9d7efdb6"
+      },
+      {
+        "case": "coding-10",
+        "options": {},
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "94354515b7bcb3d0493082d7a923fc6a3f828576f84c7851187462d32e08b4ce",
+        "token_ids_sha256": "d1006418b971215d68426eca0929c46e7e55c66c36e5bfb8f1b125297ee415ad"
+      },
+      {
+        "case": "coding-10",
+        "options": {
+          "enable_thinking": false
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "425363c857a6ab7ece15aae285272fb08356c01fe91e2e85809a337a8f7007ed",
+        "token_ids_sha256": "276077d7b9ded956051ce9ab980fefa8c761f423d570286b49defa34d1e71f92"
+      },
+      {
+        "case": "coding-10",
+        "options": {
+          "reasoning_effort": "low"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "a78a50f74c7edbdc09efa5c1ae67361be98cffdaa4836461445c57feb5bd901d",
+        "token_ids_sha256": "c6e9c7e1d58b531f655fadfa24efdd5580004bf88258d5fd93be5429873869be"
+      },
+      {
+        "case": "coding-10",
+        "options": {
+          "reasoning_effort": "medium"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "87d4ff617dfc4d3b7389dd945a6244a020ef8751963faab8d1e11eb69772eda3",
+        "token_ids_sha256": "eeb894ad3f8837c9d0cfb00d24e654f6d489b991cd4dda621e9619fa452fc2ab"
+      },
+      {
+        "case": "coding-10",
+        "options": {
+          "reasoning_effort": "xhigh"
+        },
+        "rendered_equal": true,
+        "tokens_equal": true,
+        "rendered_sha256": "94354515b7bcb3d0493082d7a923fc6a3f828576f84c7851187462d32e08b4ce",
+        "token_ids_sha256": "d1006418b971215d68426eca0929c46e7e55c66c36e5bfb8f1b125297ee415ad"
+      }
+    ],
+    "special_token_ids": {
+      "bos_token_id": {
+        "reference": null,
+        "candidate": null
+      },
+      "eos_token_id": {
+        "reference": 248046,
+        "candidate": 248046
+      },
+      "pad_token_id": {
+        "reference": 248044,
+        "candidate": 248044
+      },
+      "unk_token_id": {
+        "reference": null,
+        "candidate": null
+      }
+    },
+    "reference_commit": "5ad04445d2686f525e9fbe5c077e6fa0c7df4200",
+    "scope": "Actual tokenizer/template comparison only, no model generation"
+  },
+  "full_generation": "NOT_RUN",
+  "weight_value_equivalence_to_source": "NOT_RUN",
+  "model_quality": "NOT_RUN",
+  "script_sha256": "59470d94d54adef20c6413a9fbabdb0c19cdc6cbe225dfd760507cb28a6e906f"
+}

compatibility/patch-manifest.json

@@ -0,0 +1,13 @@
+{
+  "mlx_lm_revision": "c69d1288440a0dc4e6401fc417098b07598dccd5",
+  "patches": [
+    {
+      "path": "compatibility/swift15-mlx-lm.patch",
+      "sha256": "f6f1d0bdafa45863bfbf93dac0398c481c993ea04fdf38b9bae98c643f89eaec"
+    },
+    {
+      "path": "compatibility/enable-5bit.patch",
+      "sha256": "b985961eac3035e05ca4c9f3a8b283c26e6dd69bab4d113997fc04fe2a4f99cc"
+    }
+  ]
+}

compatibility/quant-validation-results.json

@@ -0,0 +1,103 @@
+{
+  "all_floating_tensors_finite": true,
+  "assets_sha256": {
+    "chat_template.jinja": "c3cf9e34abf4f9e36c2d72165aa9c132d3e2a725b6c2586aaa3a8af9d7a81041",
+    "generation_config.json": "e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e",
+    "merges.txt": "a9d356d7bdf1ef4949e3e748e95b8e10ad9d4e2e838eddc38a0a7b6b94d1db8d",
+    "preprocessor_config.json": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516",
+    "tokenizer.json": "0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3",
+    "tokenizer_config.json": "b11349aafa7cdc6a320767cf7ceb29ed82f7eda5d65e8e0819e76f0ce947bf27",
+    "video_preprocessor_config.json": "7768af27c1fafa9cc9011c1dc20067e03f8915e03b63504550e11d5066986d13",
+    "vocab.json": "ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003"
+  },
+  "categories": {
+    "MTP": 15,
+    "text": 851,
+    "vision": 333
+  },
+  "chat_templates": [
+    {
+      "options": {
+        "enable_thinking": false
+      },
+      "rendered": "<|im_start|>user\nSay hello.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"
+    },
+    {
+      "options": {
+        "reasoning_effort": "low"
+      },
+      "rendered": "<|im_start|>system\nReasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.<|im_end|>\n<|im_start|>user\nSay hello.<|im_end|>\n<|im_start|>assistant\n<think>\n"
+    },
+    {
+      "options": {
+        "reasoning_effort": "xhigh"
+      },
+      "rendered": "<|im_start|>system\nReasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.<|im_end|>\n<|im_start|>user\nSay hello.<|im_end|>\n<|im_start|>assistant\n<think>\n"
+    }
+  ],
+  "exact_unquantized_tensors": 609,
+  "generation": {
+    "elapsed_seconds": 495.4394222159999,
+    "finish_reason": "stop",
+    "prompt": "<|im_start|>user\nReply with exactly: Hello from Swift.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n",
+    "prompt_tokens_per_second": 0.04886516458777248,
+    "text": "Hello from Swift.",
+    "token_ids": [
+      9419,
+      494,
+      22929,
+      13,
+      248046
+    ],
+    "tokens": 5,
+    "tokens_per_second": 0.05894060179705689
+  },
+  "ignored_tensors": 0,
+  "inference_floating_dtype": "float32, CPU runtime only; stored floating tensors remain BF16",
+  "load_memory_bytes": 19281804392,
+  "load_seconds": 3.1353140849996635,
+  "mapped_source_tensors": 1199,
+  "mtp": {
+    "path": "Explicit MTP step with real text hidden states and shared LM head; speculative generation is not integrated",
+    "shape": [
+      1,
+      1,
+      248320
+    ],
+    "status": "PASS"
+  },
+  "native_bf16_cpu_inference": "Aborted after reproducing incorrect accumulation in the official Linux BF16 quantized matmul. See cpu-quantized-matmul-diagnostic.json.",
+  "process_peak_rss_bytes": 24964259840,
+  "processor": "Qwen3VLProcessor",
+  "quantization": {
+    "bits": 5,
+    "group_size": 64,
+    "mode": "affine"
+  },
+  "recorded_at": "2026-09-22T12:17:07.052752+00:00",
+  "saved_tensors": 2379,
+  "source_repo": "ukisai/Swift-1.5-Qwen3.8-27b",
+  "source_revision": "5ad04445d2686f525e9fbe5c077e6fa0c7df4200",
+  "source_shard_bytes": 55563006776,
+  "source_shards": 18,
+  "source_tensors": 1199,
+  "status": "PASS",
+  "tokenizer": "Qwen2Tokenizer",
+  "total_validation_seconds": 609.2779944080003,
+  "unexplained_tensors": 0,
+  "vision": {
+    "grid": [
+      [
+        1,
+        16,
+        16
+      ]
+    ],
+    "path": "Vision encoder only; image/video insertion and multimodal text generation are not implemented",
+    "shape": [
+      64,
+      5120
+    ],
+    "status": "PASS"
+  }
+}

compatibility/source-structural-results.json

@@ -0,0 +1,55 @@
+{
+  "categories": {
+    "MTP": 15,
+    "text": 851,
+    "vision": 333
+  },
+  "chat_templates": [
+    {
+      "options": {
+        "enable_thinking": false
+      },
+      "rendered": "<|im_start|>user\nSay hello.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n",
+      "tokens": 15
+    },
+    {
+      "options": {
+        "reasoning_effort": "low"
+      },
+      "rendered": "<|im_start|>system\nReasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.<|im_end|>\n<|im_start|>user\nSay hello.<|im_end|>\n<|im_start|>assistant\n<think>\n",
+      "tokens": 43
+    },
+    {
+      "options": {
+        "reasoning_effort": "xhigh"
+      },
+      "rendered": "<|im_start|>system\nReasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.<|im_end|>\n<|im_start|>user\nSay hello.<|im_end|>\n<|im_start|>assistant\n<think>\n",
+      "tokens": 55
+    }
+  ],
+  "elapsed_seconds": 1.0959153320000041,
+  "full_parameter_evaluation": false,
+  "generation": "NOT_RUN",
+  "ignored_tensors": 0,
+  "mapped_tensors": 1199,
+  "mlx_active_memory_bytes": 8,
+  "mtp_runtime": "component-only; no integrated speculative decoding",
+  "platform": "Linux x86_64",
+  "process_peak_rss_bytes": 1180291072,
+  "processor": "Qwen3VLProcessor",
+  "quantization_executed": false,
+  "recorded_at": "2026-09-22T12:03:20.905145+00:00",
+  "source": "<SOURCE_MODEL_DIR>",
+  "source_repo": "ukisai/Swift-1.5-Qwen3.8-27b",
+  "source_revision": "5ad04445d2686f525e9fbe5c077e6fa0c7df4200",
+  "source_tensors": 1199,
+  "source_verification": "compatibility/source-verification.json",
+  "previous_recorded_source_verification_sha256_unverified": "1e6e65ed808f3dbc08766bc141706f2d30349d28608f6aba6884b419c3849ef3",
+  "source_verification_sha256": "27d87d8d5209c6559643efdd7d7b468b0138ebfb3a7d5771b29be79214bffe75",
+  "reference_update_note": "Current hash identifies the published redacted document. The previous recorded hash is retained for provenance but was not reproduced. This reference correction does not rerun the historical validation.",
+  "status": "PASS_ACTUAL_SOURCE_LAZY_STRUCTURAL_LOAD",
+  "tokenizer": "Qwen2Tokenizer",
+  "unexplained_tensors": 0,
+  "vision_runtime": "encoder-only; no integrated multimodal generation",
+  "weight_backing": "actual verified source safetensors; lazy loading"
+}

compatibility/swift15-mlx-lm.patch

@@ -0,0 +1,962 @@
+diff --git a/mlx_lm/models/qwen3_5_full.py b/mlx_lm/models/qwen3_5_full.py
+new file mode 100644
+index 0000000..eb3d09b
+--- /dev/null
++++ b/mlx_lm/models/qwen3_5_full.py
+@@ -0,0 +1,485 @@
++# Copyright © 2026 Apple Inc.
++
++"""Complete Qwen3.5 parameter model, including the vision encoder and MTP.
++
++Text generation, the vision encoder, and explicit MTP steps are separate APIs.
++Image/video token insertion, multimodal text positions, and speculative decoding
++are not implemented here. Unsupported multimodal calls raise an error.
++"""
++
++import copy
++from dataclasses import dataclass
++from typing import Optional
++
++import mlx.core as mx
++import mlx.nn as nn
++import numpy as np
++from mlx.utils import tree_flatten, tree_unflatten
++
++from . import qwen3_5
++from .base import BaseModelArgs, create_attention_mask
++from .cache import KVCache
++
++
++class OffsetRMSNorm(nn.Module):
++    """Keep HF's zero-centered weights without rounding weight + 1 to BF16."""
++
++    def __init__(self, dims, eps=1e-6):
++        super().__init__()
++        self.weight = mx.zeros((dims,))
++        self.eps = eps
++
++    def __call__(self, x):
++        y = x.astype(mx.float32)
++        y = y * mx.rsqrt(mx.mean(y * y, axis=-1, keepdims=True) + self.eps)
++        return (y * (1 + self.weight.astype(mx.float32))).astype(x.dtype)
++
++
++def _use_offset_norms(module):
++    replacements = [
++        (name, OffsetRMSNorm(norm.weight.shape[0], norm.eps))
++        for name, norm in module.named_modules()
++        if isinstance(norm, nn.RMSNorm)
++    ]
++    module.update_modules(tree_unflatten(replacements))
++
++
++@dataclass
++class VisionArgs(BaseModelArgs):
++    depth: int
++    hidden_size: int
++    intermediate_size: int
++    num_heads: int
++    out_hidden_size: int
++    num_position_embeddings: int
++    patch_size: int = 16
++    temporal_patch_size: int = 2
++    spatial_merge_size: int = 2
++    in_channels: int = 3
++    hidden_act: str = "gelu_pytorch_tanh"
++    deepstack_visual_indexes: Optional[list] = None
++
++    def __post_init__(self):
++        if self.deepstack_visual_indexes:
++            raise ValueError("DeepStack vision features are not supported")
++        if self.hidden_act != "gelu_pytorch_tanh":
++            raise ValueError(f"Unsupported vision activation: {self.hidden_act}")
++        if self.hidden_size % self.num_heads or self.hidden_size // self.num_heads % 4:
++            raise ValueError("Vision head dimension must be divisible by four")
++        if int(self.num_position_embeddings**0.5) ** 2 != self.num_position_embeddings:
++            raise ValueError("Vision position table must be square")
++
++
++class VisionPatchEmbed(nn.Module):
++    def __init__(self, args):
++        super().__init__()
++        self.args = args
++        kernel = (args.temporal_patch_size, args.patch_size, args.patch_size)
++        self.proj = nn.Conv3d(
++            args.in_channels, args.hidden_size, kernel, stride=kernel, bias=True
++        )
++
++    def __call__(self, pixels):
++        a = self.args
++        x = pixels.reshape(
++            -1, a.in_channels, a.temporal_patch_size, a.patch_size, a.patch_size
++        )
++        x = x.transpose(0, 2, 3, 4, 1).astype(self.proj.weight.dtype)
++        return self.proj(x).reshape(-1, a.hidden_size)
++
++
++class VisionAttention(nn.Module):
++    def __init__(self, args):
++        super().__init__()
++        self.num_heads = args.num_heads
++        self.head_dim = args.hidden_size // args.num_heads
++        self.qkv = nn.Linear(args.hidden_size, 3 * args.hidden_size)
++        self.proj = nn.Linear(args.hidden_size, args.hidden_size)
++
++    def __call__(self, x, cos, sin, boundaries):
++        qkv = self.qkv(x).reshape(x.shape[0], 3, self.num_heads, self.head_dim)
++        q, k, v = qkv.transpose(1, 0, 2, 3)
++
++        def rotate(y):
++            z = y.astype(mx.float32)
++            half = self.head_dim // 2
++            rotated = mx.concatenate([-z[..., half:], z[..., :half]], axis=-1)
++            return (z * cos[:, None] + rotated * sin[:, None]).astype(y.dtype)
++
++        q, k = rotate(q), rotate(k)
++        outputs = []
++        for start, end in zip(boundaries, boundaries[1:]):
++            out = mx.fast.scaled_dot_product_attention(
++                q[start:end].transpose(1, 0, 2)[None],
++                k[start:end].transpose(1, 0, 2)[None],
++                v[start:end].transpose(1, 0, 2)[None],
++                scale=self.head_dim**-0.5,
++            )
++            outputs.append(out[0].transpose(1, 0, 2).reshape(end - start, -1))
++        return self.proj(mx.concatenate(outputs, axis=0))
++
++
++class VisionMLP(nn.Module):
++    def __init__(self, args):
++        super().__init__()
++        self.linear_fc1 = nn.Linear(args.hidden_size, args.intermediate_size)
++        self.linear_fc2 = nn.Linear(args.intermediate_size, args.hidden_size)
++
++    def __call__(self, x):
++        return self.linear_fc2(nn.gelu_approx(self.linear_fc1(x)))
++
++
++class VisionBlock(nn.Module):
++    def __init__(self, args):
++        super().__init__()
++        self.norm1 = nn.LayerNorm(args.hidden_size, eps=1e-6)
++        self.norm2 = nn.LayerNorm(args.hidden_size, eps=1e-6)
++        self.attn = VisionAttention(args)
++        self.mlp = VisionMLP(args)
++
++    def __call__(self, x, cos, sin, boundaries):
++        x = x + self.attn(self.norm1(x), cos, sin, boundaries)
++        return x + self.mlp(self.norm2(x))
++
++
++class VisionMerger(nn.Module):
++    def __init__(self, args):
++        super().__init__()
++        self.hidden_size = args.hidden_size * args.spatial_merge_size**2
++        self.norm = nn.LayerNorm(args.hidden_size, eps=1e-6)
++        self.linear_fc1 = nn.Linear(self.hidden_size, self.hidden_size)
++        self.linear_fc2 = nn.Linear(self.hidden_size, args.out_hidden_size)
++
++    def __call__(self, x):
++        x = self.norm(x).reshape(-1, self.hidden_size)
++        return self.linear_fc2(nn.gelu(self.linear_fc1(x)))
++
++
++class VisionModel(nn.Module):
++    def __init__(self, args):
++        super().__init__()
++        self.args = args
++        self.patch_embed = VisionPatchEmbed(args)
++        self.pos_embed = nn.Embedding(args.num_position_embeddings, args.hidden_size)
++        self.blocks = [VisionBlock(args) for _ in range(args.depth)]
++        self.merger = VisionMerger(args)
++
++    def _positions(self, grid_thw):
++        grid = np.asarray(
++            grid_thw.tolist() if hasattr(grid_thw, "tolist") else grid_thw
++        )
++        if (
++            grid.ndim != 2
++            or grid.shape[1] != 3
++            or not np.issubdtype(grid.dtype, np.integer)
++        ):
++            raise ValueError("grid_thw must be an integer array with shape (N, 3)")
++        merge = self.args.spatial_merge_size
++        side = int(self.args.num_position_embeddings**0.5)
++        positions, indices, weights, boundaries = [], [], [], [0]
++        for t, h, w in grid.tolist():
++            if min(t, h, w) <= 0 or h % merge or w % merge:
++                raise ValueError("Invalid vision grid or spatial merge dimensions")
++            hp, wp = np.indices((h, w))
++            block = (h // merge, merge, w // merge, merge)
++            hp = hp.reshape(block).transpose(0, 2, 1, 3).reshape(-1)
++            wp = wp.reshape(block).transpose(0, 2, 1, 3).reshape(-1)
++            positions.append(np.tile(np.stack([hp, wp], axis=-1), (t, 1)))
++            reorder = np.tile(hp * w + wp, t)
++            hs = np.linspace(0, side - 1, h, dtype=np.float32)
++            ws = np.linspace(0, side - 1, w, dtype=np.float32)
++            hf, wf = hs.astype(np.int32), ws.astype(np.int32)
++            hc, wc = np.minimum(hf + 1, side - 1), np.minimum(wf + 1, side - 1)
++            dh, dw = hs - hf, ws - wf
++            idx = [
++                (a[:, None] * side + b[None]).reshape(-1)
++                for a, b in [(hf, wf), (hf, wc), (hc, wf), (hc, wc)]
++            ]
++            coeff = [
++                (a[:, None] * b[None]).reshape(-1)
++                for a, b in [(1 - dh, 1 - dw), (1 - dh, dw), (dh, 1 - dw), (dh, dw)]
++            ]
++            indices.append(np.stack(idx)[:, reorder])
++            weights.append(np.stack(coeff)[:, reorder])
++            offset = boundaries[-1]
++            boundaries.extend(offset + (i + 1) * h * w for i in range(t))
++        if not positions:
++            raise ValueError("At least one vision grid is required")
++        return (
++            mx.array(np.concatenate(positions), dtype=mx.float32),
++            mx.array(np.concatenate(indices, axis=1), dtype=mx.int32),
++            mx.array(np.concatenate(weights, axis=1), dtype=mx.float32),
++            boundaries,
++        )
++
++    def __call__(self, pixels, grid_thw, return_hidden_states=False):
++        positions, indices, weights, boundaries = self._positions(grid_thw)
++        x = self.patch_embed(pixels)
++        if x.shape[0] != boundaries[-1]:
++            raise ValueError("Pixel patch count does not match grid_thw")
++        pos = (self.pos_embed(indices) * weights[..., None]).sum(axis=0)
++        x = x + pos.astype(x.dtype)
++        dim = self.args.hidden_size // self.args.num_heads // 2
++        inv_freq = 1.0 / (10000 ** (mx.arange(0, dim, 2, dtype=mx.float32) / dim))
++        angles = (positions[..., None] * inv_freq).reshape(x.shape[0], -1)
++        angles = mx.concatenate([angles, angles], axis=-1)
++        cos, sin = mx.cos(angles), mx.sin(angles)
++        for block in self.blocks:
++            x = block(x, cos, sin, boundaries)
++        merged = self.merger(x)
++        return (x, merged) if return_hidden_states else merged
++
++
++class MultiTokenPredictor(nn.Module):
++    """An explicit MTP step; embeddings and the output head belong to the LM."""
++
++    def __init__(self, args, num_layers):
++        super().__init__()
++        self.fc = nn.Linear(2 * args.hidden_size, args.hidden_size, bias=False)
++        self.pre_fc_norm_embedding = OffsetRMSNorm(args.hidden_size, args.rms_norm_eps)
++        self.pre_fc_norm_hidden = OffsetRMSNorm(args.hidden_size, args.rms_norm_eps)
++        self.layers = [
++            qwen3_5.DecoderLayer(args, args.full_attention_interval - 1)
++            for _ in range(num_layers)
++        ]
++        self.norm = OffsetRMSNorm(args.hidden_size, args.rms_norm_eps)
++        _use_offset_norms(self)
++
++    def __call__(self, hidden_states, next_token_embeddings, cache=None, step=0):
++        if hidden_states.shape != next_token_embeddings.shape:
++            raise ValueError("MTP hidden states and next-token embeddings must align")
++        if step < 0:
++            raise ValueError("MTP step must be non-negative")
++        x = mx.concatenate(
++            [
++                self.pre_fc_norm_embedding(next_token_embeddings),
++                self.pre_fc_norm_hidden(hidden_states),
++            ],
++            axis=-1,
++        )
++        x = self.fc(x)
++        layer = step % len(self.layers)
++        if cache is not None and len(cache) != len(self.layers):
++            raise ValueError("MTP requires one KV cache per MTP layer")
++        c = cache[layer] if cache is not None else None
++        return self.norm(self.layers[layer](x, create_attention_mask(x, c), c))
++
++    def make_cache(self):
++        return [KVCache() for _ in self.layers]
++
++
++@dataclass
++class ModelArgs(BaseModelArgs):
++    model_type: str
++    text_config: dict
++    vision_config: dict
++    language_model_only: bool = False
++    image_token_id: Optional[int] = None
++    video_token_id: Optional[int] = None
++    vision_start_token_id: Optional[int] = None
++    tie_word_embeddings: bool = False
++    quantization: Optional[dict] = None
++
++    @classmethod
++    def from_dict(cls, params):
++        return super().from_dict(copy.deepcopy(params))
++
++
++class Model(nn.Module):
++    extra_save_files = (
++        "preprocessor_config.json",
++        "video_preprocessor_config.json",
++        "processor_config.json",
++        "tokenizer.json",
++        "tokenizer_config.json",
++        "special_tokens_map.json",
++        "vocab.json",
++        "merges.txt",
++        "chat_template.jinja",
++    )
++
++    def __init__(self, args):
++        super().__init__()
++        self.args = args
++        self.model_type = args.model_type
++        text = args.text_config
++        if args.language_model_only:
++            raise ValueError("The complete model requires language_model_only=false")
++        if text.get("hidden_act", "silu") not in ("silu", "swish"):
++            raise ValueError("Unsupported text MLP activation")
++        if text.get("attn_output_gate", True) is not True:
++            raise ValueError("Ungated attention is not supported")
++        if text.get("output_gate_type", "swish") not in ("swish", "sigmoid"):
++            raise ValueError("Unknown attention gate declaration")
++        if text.get("mtp_use_dedicated_embeddings", False):
++            raise ValueError("Dedicated MTP embeddings are not supported")
++        if args.tie_word_embeddings != text.get("tie_word_embeddings", False):
++            raise ValueError("Conflicting text and top-level weight tying settings")
++        self._text_args = qwen3_5.TextModelArgs.from_dict(copy.deepcopy(text))
++        expected_layers = [
++            (
++                "full_attention"
++                if (i + 1) % self._text_args.full_attention_interval == 0
++                else "linear_attention"
++            )
++            for i in range(self._text_args.num_hidden_layers)
++        ]
++        if text.get("layer_types", expected_layers) != expected_layers:
++            raise ValueError("Layer schedule differs from the supported architecture")
++        if self._text_args.num_experts:
++            raise ValueError("This complete model supports the dense architecture")
++        self.language_model = qwen3_5.TextModel(self._text_args)
++        _use_offset_norms(self.language_model)
++        self.visual = VisionModel(VisionArgs.from_dict(args.vision_config))
++        if self.visual.args.out_hidden_size != self._text_args.hidden_size:
++            raise ValueError("Vision output size does not match text hidden size")
++        num_mtp = text.get("mtp_num_hidden_layers", 0)
++        if num_mtp < 0:
++            raise ValueError("Invalid MTP layer count")
++        if num_mtp:
++            self.mtp = MultiTokenPredictor(self._text_args, num_mtp)
++        self._parameter_shapes = {
++            k: tuple(v.shape) for k, v in tree_flatten(self.parameters())
++        }
++
++    @property
++    def model(self):
++        return self.language_model.model
++
++    @property
++    def layers(self):
++        return self.language_model.layers
++
++    def make_cache(self):
++        return self.language_model.make_cache()
++
++    def __call__(self, inputs, cache=None, input_embeddings=None, **kwargs):
++        if kwargs:
++            raise NotImplementedError(
++                "Multimodal text integration is not implemented; use visual() for encoder features"
++            )
++        for token in (
++            self.args.image_token_id,
++            self.args.video_token_id,
++            self.args.vision_start_token_id,
++        ):
++            if (
++                token is not None
++                and inputs is not None
++                and bool(mx.any(inputs == token))
++            ):
++                raise NotImplementedError(
++                    "Multimodal token positions require a multimodal text runtime"
++                )
++        return self.language_model(inputs, cache, input_embeddings)
++
++    def mtp_logits(self, next_token_ids, previous_hidden_states, cache=None, step=0):
++        if not hasattr(self, "mtp"):
++            raise ValueError("This checkpoint has no MTP layers")
++        embeddings = self.model.embed_tokens(next_token_ids)
++        hidden = self.mtp(previous_hidden_states, embeddings, cache, step)
++        if self._text_args.tie_word_embeddings:
++            return self.model.embed_tokens.as_linear(hidden)
++        return self.language_model.lm_head(hidden)
++
++    def weight_mapping(self):
++        rows = []
++        for name, shape in sorted(self._parameter_shapes.items()):
++            source, source_shape, transform = name, shape, "identity"
++            if name.startswith("language_model.model."):
++                source = "model.language_model." + name[len("language_model.model.") :]
++            elif name.startswith("language_model.lm_head."):
++                source = name[len("language_model.") :]
++            elif name.startswith("visual."):
++                source = "model." + name
++            if name.endswith(".conv1d.weight"):
++                source_shape = (shape[0], shape[2], shape[1])
++                transform = "transpose(0,2,1)"
++            elif name == "visual.patch_embed.proj.weight":
++                source_shape = (shape[0], shape[4], shape[1], shape[2], shape[3])
++                transform = "transpose(0,2,3,4,1)"
++            category = (
++                "vision"
++                if name.startswith("visual.")
++                else "MTP" if name.startswith("mtp.") else "text"
++            )
++            rows.append(
++                dict(
++                    source=source,
++                    source_shape=source_shape,
++                    destination=name,
++                    destination_shape=shape,
++                    category=category,
++                    transform=transform,
++                )
++            )
++        return rows
++
++    def sanitize(self, weights):
++        rows = self.weight_mapping()
++        hf_names = {r["source"] for r in rows}
++        native_names = set(self._parameter_shapes)
++        is_hf = any(k.startswith("model.language_model.") for k in weights)
++        expected = hf_names if is_hf else native_names
++        if self.args.quantization:
++            if is_hf:
++                raise ValueError("Only original unquantized HF weights can be mapped")
++            return self._check_native_quantized(weights)
++        missing, unexpected = expected - set(weights), set(weights) - expected
++        if missing or unexpected:
++            raise ValueError(
++                f"Incomplete weight mapping: missing={sorted(missing)}, unexpected={sorted(unexpected)}"
++            )
++        mapped = {}
++        for row in rows:
++            key = row["source"] if is_hf else row["destination"]
++            value = weights[key]
++            shape = row["source_shape"] if is_hf else row["destination_shape"]
++            if tuple(value.shape) != shape:
++                raise ValueError(f"Shape mismatch for {key}: {value.shape} != {shape}")
++            if is_hf and row["transform"] == "transpose(0,2,1)":
++                value = value.transpose(0, 2, 1)
++            elif is_hf and row["transform"] == "transpose(0,2,3,4,1)":
++                value = value.transpose(0, 2, 3, 4, 1)
++            mapped[row["destination"]] = value
++        return mapped
++
++    def _check_native_quantized(self, weights):
++        shapes = dict(self._parameter_shapes)
++        q = self.args.quantization
++        for path, module in self.named_modules():
++            if not hasattr(module, "to_quantized"):
++                continue
++            settings = q.get(path, q)
++            if settings is False:
++                continue
++            if not isinstance(settings, dict):
++                raise ValueError(f"Invalid native quantization metadata for {path}")
++            bits, group, mode = (
++                settings.get("bits"),
++                settings.get("group_size"),
++                settings.get("mode", "affine"),
++            )
++            if (bits, group, mode) != (4, 64, "affine"):
++                raise ValueError(
++                    "This extension only prepares affine/4-bit/group-64 native checkpoints"
++                )
++            original = shapes[f"{path}.weight"]
++            if original[-1] % group:
++                continue
++            shapes[f"{path}.weight"] = (*original[:-1], original[-1] * bits // 32)
++            shapes[f"{path}.scales"] = (*original[:-1], original[-1] // group)
++            shapes[f"{path}.biases"] = shapes[f"{path}.scales"]
++        missing, unexpected = set(shapes) - set(weights), set(weights) - set(shapes)
++        if missing or unexpected:
++            raise ValueError(
++                f"Incomplete native checkpoint: missing={sorted(missing)}, unexpected={sorted(unexpected)}"
++            )
++        for name, shape in shapes.items():
++            if tuple(weights[name].shape) != shape:
++                raise ValueError(f"Native checkpoint shape mismatch: {name}")
++        return weights
++
++    @property
++    def cast_predicate(self):
++        return self.language_model.cast_predicate
+diff --git a/mlx_lm/utils.py b/mlx_lm/utils.py
+index a00fed8..6a751a2 100644
+--- a/mlx_lm/utils.py
++++ b/mlx_lm/utils.py
+@@ -195,6 +195,13 @@ def _transform_awq_weights(
+     return new_weights, mlx_quantization
+ 
+ 
++def _is_complete_qwen3_5(config: dict) -> bool:
++    return (
++        config.get("model_type") == "qwen3_5"
++        and config.get("language_model_only") is False
++    )
++
++
+ def _get_classes(config: dict):
+     """
+     Retrieve the model and model args classes based on the configuration.
+@@ -216,6 +223,8 @@ def _get_classes(config: dict):
+             break
+     else:
+         model_type = MODEL_REMAPPING.get(model_type, model_type)
++    if _is_complete_qwen3_5(config):
++        model_type = "qwen3_5_full"
+     try:
+         arch = importlib.import_module(f"mlx_lm.models.{model_type}")
+     except ImportError as e:
+@@ -446,12 +455,35 @@ def load_model(
+ 
+     weight_files = glob.glob(str(model_path / "model*.safetensors"))
+ 
++    complete_qwen = _is_complete_qwen3_5(config)
++    if complete_qwen:
++        with open(model_path / "model.safetensors.index.json") as stream:
++            index = json.load(stream)
++        expected_files = set(index["weight_map"].values())
++        actual_files = {Path(file).name for file in weight_files}
++        if actual_files != expected_files:
++            raise ValueError(
++                "Incomplete full-model checkpoint: "
++                f"missing shards={sorted(expected_files - actual_files)}, "
++                f"unexpected shards={sorted(actual_files - expected_files)}"
++            )
++
+     if not weight_files and strict:
+         raise FileNotFoundError(f"No safetensors found in {model_path}")
+ 
+     weights = {}
+     for wf in weight_files:
+-        weights.update(mx.load(wf))
++        shard = mx.load(wf)
++        if complete_qwen:
++            duplicates = weights.keys() & shard.keys()
++            if duplicates:
++                raise ValueError(f"Duplicate checkpoint tensors: {sorted(duplicates)}")
++            for name in shard:
++                if index["weight_map"].get(name) != Path(wf).name:
++                    raise ValueError(f"Checkpoint index mismatch: {name}")
++        weights.update(shard)
++    if complete_qwen and weights.keys() != index["weight_map"].keys():
++        raise ValueError("Checkpoint tensor set does not match its index")
+ 
+     if (model_file := config.get("model_file")) is not None:
+         if not trust_remote_code:
+@@ -1064,9 +1096,11 @@ def save_config(
+         config (dict): The model configuration.
+         config_path (Union[str, Path]): Model configuration file path.
+     """
+-    # Clean unused keys
++    config = copy.deepcopy(config)
++    # Complete multimodal checkpoints need the vision architecture on reload.
+     config.pop("_name_or_path", None)
+-    config.pop("vision_config", None)
++    if not _is_complete_qwen3_5(config):
++        config.pop("vision_config", None)
+     if "quantization" in config:
+         config["quantization_config"] = config["quantization"]
+ 
+@@ -1099,7 +1133,8 @@ def save(
+     save_config(config, config_path=dst_path / "config.json")
+     tokenizer.save_pretrained(dst_path)
+ 
+-    for p in ["*.py", "generation_config.json"]:
++    extra_files = getattr(model, "extra_save_files", ())
++    for p in ["*.py", "generation_config.json", *extra_files]:
+         for file in glob.glob(str(src_path / p)):
+             shutil.copy(file, dst_path)
+ 
+diff --git a/tests/test_qwen3_5_full.py b/tests/test_qwen3_5_full.py
+new file mode 100644
+index 0000000..2b3e5ae
+--- /dev/null
++++ b/tests/test_qwen3_5_full.py
+@@ -0,0 +1,377 @@
++# Copyright © 2026 Apple Inc.
++
++"""Small synthetic fixtures test architecture code, not Swift model quality."""
++
++import copy
++import json
++from pathlib import Path
++from unittest.mock import patch
++
++import mlx.core as mx
++import numpy as np
++import pytest
++import torch
++from mlx.utils import tree_flatten
++from mlx_lm.convert import convert
++from mlx_lm.models.qwen3_5_full import Model, ModelArgs, OffsetRMSNorm
++from mlx_lm.utils import _get_classes, load_model, save_config
++from transformers import Qwen3_5Config, Qwen3_5ForConditionalGeneration
++from transformers.models.qwen3_5.modeling_qwen3_5 import (
++    Qwen3_5DecoderLayer,
++    Qwen3_5RMSNorm,
++    Qwen3_5TextRotaryEmbedding,
++)
++
++
++def small_config():
++    return {
++        "model_type": "qwen3_5",
++        "architectures": ["Qwen3_5ForConditionalGeneration"],
++        "language_model_only": False,
++        "tie_word_embeddings": False,
++        "image_token_id": 125,
++        "video_token_id": 126,
++        "vision_start_token_id": 127,
++        "text_config": {
++            "model_type": "qwen3_5_text",
++            "hidden_size": 128,
++            "intermediate_size": 256,
++            "num_hidden_layers": 4,
++            "num_attention_heads": 4,
++            "num_key_value_heads": 2,
++            "head_dim": 32,
++            "vocab_size": 128,
++            "full_attention_interval": 4,
++            "layer_types": ["linear_attention"] * 3 + ["full_attention"],
++            "linear_num_key_heads": 2,
++            "linear_num_value_heads": 4,
++            "linear_key_head_dim": 32,
++            "linear_value_head_dim": 32,
++            "linear_conv_kernel_dim": 4,
++            "hidden_act": "silu",
++            "attn_output_gate": True,
++            "output_gate_type": "swish",
++            "mamba_ssm_dtype": "float32",
++            "rms_norm_eps": 1e-6,
++            "max_position_embeddings": 256,
++            "tie_word_embeddings": False,
++            "attention_bias": False,
++            "attention_dropout": 0.0,
++            "mtp_num_hidden_layers": 1,
++            "mtp_use_dedicated_embeddings": False,
++            "rope_parameters": {
++                "rope_type": "default",
++                "rope_theta": 10000000,
++                "partial_rotary_factor": 0.5,
++                "mrope_interleaved": True,
++                "mrope_section": [3, 3, 2],
++            },
++        },
++        "vision_config": {
++            "model_type": "qwen3_5",
++            "depth": 2,
++            "hidden_size": 32,
++            "intermediate_size": 48,
++            "num_heads": 4,
++            "out_hidden_size": 128,
++            "num_position_embeddings": 16,
++            "patch_size": 2,
++            "temporal_patch_size": 2,
++            "spatial_merge_size": 2,
++            "in_channels": 3,
++            "hidden_act": "gelu_pytorch_tanh",
++            "deepstack_visual_indexes": [],
++        },
++    }
++
++
++class ReferenceMTP(torch.nn.Module):
++    """Single-step composition used by the source vLLM MTP implementation."""
++
++    def __init__(self, args):
++        super().__init__()
++        h = args.hidden_size
++        self.fc = torch.nn.Linear(2 * h, h, bias=False)
++        self.pre_fc_norm_embedding = Qwen3_5RMSNorm(h, args.rms_norm_eps)
++        self.pre_fc_norm_hidden = Qwen3_5RMSNorm(h, args.rms_norm_eps)
++        self.layers = torch.nn.ModuleList([Qwen3_5DecoderLayer(args, 3)])
++        self.norm = Qwen3_5RMSNorm(h, args.rms_norm_eps)
++        self.rotary = Qwen3_5TextRotaryEmbedding(args)
++
++    def forward(self, hidden, embeds):
++        x = self.fc(
++            torch.cat(
++                [self.pre_fc_norm_embedding(embeds), self.pre_fc_norm_hidden(hidden)],
++                dim=-1,
++            )
++        )
++        positions = torch.arange(x.shape[1])[None, None].expand(3, x.shape[0], -1)
++        rotary = self.rotary(x, positions)
++        mask = torch.triu(
++            torch.full((x.shape[0], 1, x.shape[1], x.shape[1]), float("-inf")),
++            diagonal=1,
++        )
++        return self.norm(
++            self.layers[0](x, position_embeddings=rotary, attention_mask=mask)
++        )
++
++
++@pytest.fixture(scope="module")
++def reference():
++    torch.set_num_threads(2)
++    torch.manual_seed(71)
++    config = small_config()
++    hf_config = Qwen3_5Config(**copy.deepcopy(config))
++    hf_config._attn_implementation = "eager"
++    hf_config.text_config._attn_implementation = "eager"
++    hf_config.vision_config._attn_implementation = "eager"
++    hf = Qwen3_5ForConditionalGeneration(hf_config).eval()
++    mtp = ReferenceMTP(hf_config.text_config).eval()
++    with torch.no_grad():
++        for module in (hf, mtp):
++            for name, value in module.named_parameters():
++                if "norm" in name and value.ndim == 1 and "model.visual" not in name:
++                    value.uniform_(-0.15, 0.15)
++    state = {k: mx.array(v.detach().numpy()) for k, v in hf.state_dict().items()}
++    state.update(
++        {"mtp." + k: mx.array(v.detach().numpy()) for k, v in mtp.state_dict().items()}
++    )
++    model = Model(ModelArgs.from_dict(config))
++    model.load_weights(list(model.sanitize(state).items()), strict=True)
++    model.eval()
++    return config, hf, mtp, state, model
++
++
++def assert_close(mlx_value, torch_value, atol=3e-5, rtol=3e-5):
++    np.testing.assert_allclose(
++        np.array(mlx_value), torch_value.detach().numpy(), atol=atol, rtol=rtol
++    )
++
++
++def test_dispatch_and_mapping_preserve_every_component(reference):
++    config, _, _, state, model = reference
++    before = copy.deepcopy(config)
++    cls, args = _get_classes(config)
++    fresh = cls(args.from_dict(config))
++    assert cls is Model
++    assert config == before
++    rows = fresh.weight_mapping()
++    assert {r["source"] for r in rows} == set(state)
++    assert len(rows) == len(dict(tree_flatten(model.parameters())))
++    assert sum(r["category"] == "MTP" for r in rows) == 15
++    assert sum(r["category"] == "vision" for r in rows) == 33
++    assert model.language_model.lm_head is not model.model.embed_tokens
++    assert not any("mtp.embed" in r["destination"] for r in rows)
++
++
++@pytest.mark.parametrize("prefix", ["model.language_model.", "model.visual.", "mtp."])
++def test_missing_critical_tensor_is_rejected(reference, prefix):
++    _, _, _, state, model = reference
++    missing = dict(state)
++    missing.pop(next(k for k in state if k.startswith(prefix)))
++    with pytest.raises(ValueError, match="missing="):
++        model.sanitize(missing)
++
++
++def test_unexpected_and_misshaped_tensors_are_rejected(reference):
++    _, _, _, state, model = reference
++    with pytest.raises(ValueError, match="unexpected="):
++        model.sanitize(dict(state, **{"mtp.unknown.weight": mx.zeros((1,))}))
++    wrong = dict(state)
++    wrong["mtp.fc.weight"] = mx.zeros((1,))
++    with pytest.raises(ValueError, match="Shape mismatch"):
++        model.sanitize(wrong)
++
++
++def test_native_mapping_is_idempotent_and_norm_values_are_exact(reference):
++    _, _, _, state, model = reference
++    once = model.sanitize(state)
++    twice = model.sanitize(once)
++    for key in once:
++        np.testing.assert_array_equal(np.array(once[key]), np.array(twice[key]))
++    for row in model.weight_mapping():
++        if "norm" in row["source"]:
++            np.testing.assert_array_equal(
++                np.array(state[row["source"]]), np.array(once[row["destination"]])
++            )
++
++
++def test_offset_norm_preserves_small_bf16_deltas():
++    layer = OffsetRMSNorm(4)
++    raw = mx.array([0.0001, -0.0002, 0.0003, -0.0004], dtype=mx.bfloat16)
++    layer.weight = raw
++    x = mx.array([[0.8, -1.2, 0.4, 2.0]], dtype=mx.bfloat16)
++    torch_layer = Qwen3_5RMSNorm(4)
++    torch_layer.weight.data.copy_(torch.tensor(np.array(raw.astype(mx.float32))))
++    expected = torch_layer(
++        torch.tensor(np.array(x.astype(mx.float32))).to(torch.bfloat16)
++    ).float()
++    assert_close(layer(x).astype(mx.float32), expected, atol=0, rtol=0)
++    np.testing.assert_array_equal(
++        np.array(layer.weight.astype(mx.float32)), np.array(raw.astype(mx.float32))
++    )
++
++
++def test_text_forward_matches_transformers(reference):
++    _, hf, _, _, model = reference
++    ids = torch.tensor([[3, 19, 8, 21, 11]])
++    with torch.no_grad():
++        expected = hf(input_ids=ids, use_cache=False).logits
++    actual = model(mx.array(ids.numpy()))
++    assert_close(actual, expected)
++
++
++def test_text_decode_cache_matches_full_forward(reference):
++    _, _, _, _, model = reference
++    ids = mx.array([[3, 19, 8, 21, 11]])
++    full = model(ids)
++    cache = model.make_cache()
++    parts = [model(ids[:, :3], cache)]
++    parts.extend(model(ids[:, i : i + 1], cache) for i in range(3, 5))
++    np.testing.assert_allclose(
++        np.array(mx.concatenate(parts, axis=1)), np.array(full), atol=3e-5, rtol=3e-5
++    )
++
++
++@pytest.mark.parametrize("grid", [[[1, 4, 4]], [[2, 2, 4], [1, 4, 2]], [[1, 6, 4]]])
++def test_vision_encoder_matches_transformers(reference, grid):
++    _, hf, _, _, model = reference
++    torch.manual_seed(11)
++    count = sum(t * h * w for t, h, w in grid)
++    pixels = torch.randn(count, 3 * 2 * 2 * 2)
++    with torch.no_grad():
++        expected = hf.model.visual(pixels, torch.tensor(grid))
++    hidden, pooled = model.visual(
++        mx.array(pixels.numpy()), grid, return_hidden_states=True
++    )
++    assert_close(hidden, expected.last_hidden_state)
++    assert_close(pooled, expected.pooler_output)
++
++
++def test_mtp_forward_and_shared_head_match_reference(reference):
++    _, hf, mtp, _, model = reference
++    torch.manual_seed(9)
++    hidden = torch.randn(1, 4, 128)
++    ids = torch.tensor([[9, 8, 7, 6]])
++    with torch.no_grad():
++        embeds = hf.model.language_model.embed_tokens(ids)
++        expected = mtp(hidden, embeds)
++        expected_logits = hf.lm_head(expected)
++    assert_close(
++        model.mtp(mx.array(hidden.numpy()), mx.array(embeds.numpy())), expected
++    )
++    assert_close(
++        model.mtp_logits(mx.array(ids.numpy()), mx.array(hidden.numpy())),
++        expected_logits,
++    )
++    cache = model.mtp.make_cache()
++    cached = mx.concatenate(
++        [
++            model.mtp(
++                mx.array(hidden[:, i : i + 1].numpy()),
++                mx.array(embeds[:, i : i + 1].numpy()),
++                cache,
++            )
++            for i in range(4)
++        ],
++        axis=1,
++    )
++    assert_close(cached, expected)
++
++
++def test_unsupported_multimodal_generation_fails_explicitly(reference):
++    _, _, _, _, model = reference
++    with pytest.raises(NotImplementedError, match="Multimodal"):
++        model(mx.array([[1, 2]]), pixel_values=mx.zeros((1, 24)))
++    with pytest.raises(NotImplementedError, match="Multimodal"):
++        model(mx.array([[1, 125]]))
++
++
++def test_save_config_keeps_vision_and_does_not_mutate_input(tmp_path):
++    config = small_config()
++    before = copy.deepcopy(config)
++    save_config(config, tmp_path / "config.json")
++    assert config == before
++    assert json.loads((tmp_path / "config.json").read_text()) == before
++
++
++def test_official_nonquantized_convert_and_reload_preserve_all_tensors(
++    reference, tmp_path
++):
++    config, _, _, state, model = reference
++    source, output = tmp_path / "hf", tmp_path / "mlx"
++    source.mkdir()
++    (source / "config.json").write_text(json.dumps(config))
++    mx.save_safetensors(str(source / "model.safetensors"), state)
++    (source / "model.safetensors.index.json").write_text(
++        json.dumps({"weight_map": {k: "model.safetensors" for k in state}})
++    )
++    assets = ["generation_config.json", *model.extra_save_files]
++    for name in assets:
++        (source / name).write_text("original synthetic asset: " + name)
++    (source / "generation_config.json").write_text('{"eos_token_id": 2}')
++
++    class FixtureTokenizer:
++        def save_pretrained(self, target):
++            Path(target, "tokenizer_config.json").write_text("rewritten")
++
++    def fixture_load(path, **kwargs):
++        loaded, loaded_config = load_model(Path(path), lazy=True, strict=True)
++        return loaded, FixtureTokenizer(), loaded_config
++
++    with patch("mlx_lm.convert.load", side_effect=fixture_load):
++        convert(str(source), str(output), quantize=False)
++    loaded, saved_config = load_model(output, lazy=False, strict=True)
++    expected = model.sanitize(state)
++    actual = dict(tree_flatten(loaded.parameters()))
++    assert set(actual) == set(expected)
++    for name in actual:
++        np.testing.assert_array_equal(np.array(actual[name]), np.array(expected[name]))
++    assert saved_config["vision_config"] == config["vision_config"]
++    assert saved_config["text_config"] == config["text_config"]
++    assert "quantization" not in saved_config
++    for name in assets:
++        assert (source / name).read_bytes() == (output / name).read_bytes()
++    ids = mx.array([[3, 19, 8]])
++    np.testing.assert_array_equal(np.array(model(ids)), np.array(loaded(ids)))
++    (output / "model.safetensors.index.json").write_text(
++        json.dumps(
++            {
++                "weight_map": {
++                    **{k: "model.safetensors" for k in actual},
++                    "mtp.missing.weight": "missing.safetensors",
++                }
++            }
++        )
++    )
++    with pytest.raises(ValueError, match="missing shards"):
++        load_model(output, lazy=True)
++
++
++def test_fixed_affine_quantized_roundtrip_preserves_component_tree(reference, tmp_path):
++    from mlx_lm.utils import quantize_model, save_model
++
++    config, _, _, state, _ = reference
++    model = Model(ModelArgs.from_dict(config))
++    model.load_weights(list(model.sanitize(state).items()), strict=True)
++    original_names = set(dict(tree_flatten(model.parameters())))
++    model, quantized_config = quantize_model(model, config, 64, 4, mode="affine")
++    save_model(tmp_path, model)
++    save_config(quantized_config, tmp_path / "config.json")
++    loaded, saved_config = load_model(tmp_path, lazy=False, strict=True)
++    actual = dict(tree_flatten(loaded.parameters()))
++    expected = dict(tree_flatten(model.parameters()))
++    assert set(actual) == set(expected)
++    assert original_names <= set(actual)
++    assert saved_config["quantization"] == {
++        "bits": 4,
++        "group_size": 64,
++        "mode": "affine",
++    }
++    for name in actual:
++        np.testing.assert_array_equal(np.array(actual[name]), np.array(expected[name]))
++    assert bool(mx.all(mx.isfinite(loaded(mx.array([[3, 19, 8]])))))
++    missing = dict(actual)
++    missing.pop("mtp.fc.scales")
++    with pytest.raises(ValueError, match="Incomplete native checkpoint"):
++        loaded.sanitize(missing)

config.json

@@ -0,0 +1,154 @@
+{
+    "architectures": [
+        "Qwen3_5ForConditionalGeneration"
+    ],
+    "eos_token_id": [
+        248046,
+        248044
+    ],
+    "image_token_id": 248056,
+    "language_model_only": false,
+    "model_type": "qwen3_5",
+    "quantization": {
+        "group_size": 64,
+        "bits": 5,
+        "mode": "affine"
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 5,
+        "mode": "affine"
+    },
+    "text_config": {
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attn_output_gate": true,
+        "bos_token_id": 248044,
+        "dtype": "bfloat16",
+        "eos_token_id": 248044,
+        "full_attention_interval": 4,
+        "head_dim": 256,
+        "hidden_act": "silu",
+        "hidden_size": 5120,
+        "initializer_range": 0.02,
+        "intermediate_size": 17408,
+        "layer_types": [
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention",
+            "linear_attention",
+            "linear_attention",
+            "linear_attention",
+            "full_attention"
+        ],
+        "linear_conv_kernel_dim": 4,
+        "linear_key_head_dim": 128,
+        "linear_num_key_heads": 16,
+        "linear_num_value_heads": 48,
+        "linear_value_head_dim": 128,
+        "mamba_ssm_dtype": "float32",
+        "max_position_embeddings": 262144,
+        "model_type": "qwen3_5_text",
+        "mtp_num_hidden_layers": 1,
+        "mtp_use_dedicated_embeddings": false,
+        "num_attention_heads": 24,
+        "num_hidden_layers": 64,
+        "num_key_value_heads": 4,
+        "output_gate_type": "swish",
+        "pad_token_id": null,
+        "partial_rotary_factor": 0.25,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "mrope_interleaved": true,
+            "mrope_section": [
+                11,
+                11,
+                10
+            ],
+            "partial_rotary_factor": 0.25,
+            "rope_theta": 10000000,
+            "rope_type": "default"
+        },
+        "tie_word_embeddings": false,
+        "use_cache": true,
+        "vocab_size": 248320
+    },
+    "tie_word_embeddings": false,
+    "transformers_version": "5.8.0.dev0",
+    "video_token_id": 248057,
+    "vision_config": {
+        "deepstack_visual_indexes": [],
+        "depth": 27,
+        "hidden_act": "gelu_pytorch_tanh",
+        "hidden_size": 1152,
+        "in_channels": 3,
+        "initializer_range": 0.02,
+        "intermediate_size": 4304,
+        "model_type": "qwen3_5",
+        "num_heads": 16,
+        "num_position_embeddings": 2304,
+        "out_hidden_size": 5120,
+        "patch_size": 16,
+        "spatial_merge_size": 2,
+        "temporal_patch_size": 2
+    },
+    "vision_end_token_id": 248054,
+    "vision_start_token_id": 248053
+}

generation_config.json

@@ -0,0 +1,12 @@
+{
+    "bos_token_id": 248044,
+    "do_sample": true,
+    "eos_token_id": [
+        248046,
+        248044
+    ],
+    "pad_token_id": 248044,
+    "temperature": 1.0,
+    "top_k": 20,
+    "top_p": 0.95
+}

model-00001-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9da8750595ec42d8c5d2ff0ae0b695e387cbcb3c0beac40f262e6d9a7196b59d
+size 5360417859

model-00002-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b592723bd4e9f01344335cd7d4b4f41a7c8cde98e2e074ad1d943cdb951a6835
+size 5338181777

model-00003-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0027e7fe05310d5b2d960414221ed0872de9b511aa7e6b1c3a8f0e4647558412
+size 5355925620

model-00004-of-00004.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afec577f16805b327471561d6841662eb4acf7d866ad5e761c5ef8cd3149d8f4
+size 3227578201

preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 16777216,
+        "shortest_edge": 65536
+    },
+    "patch_size": 16,
+    "temporal_patch_size": 2,
+    "merge_size": 2,
+    "image_mean": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "image_std": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "processor_class": "Qwen3VLProcessor",
+    "image_processor_type": "Qwen2VLImageProcessorFast"
+}

swift-1.5-planet-demo.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cda6924169e8b83e5d6d7299baf8329ed1b1a182a481aa0e06344e0f5c00111
+size 4738825

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0997f410c57a1f4e53b09e4be8f4a172d90edd9564368fb0847030937229b9f3
+size 12809320

tokenizer_config.json

@@ -0,0 +1,305 @@
+{
+    "add_prefix_space": false,
+    "added_tokens_decoder": {
+        "248044": {
+            "content": "<|endoftext|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248045": {
+            "content": "<|im_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248046": {
+            "content": "<|im_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248047": {
+            "content": "<|object_ref_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248048": {
+            "content": "<|object_ref_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248049": {
+            "content": "<|box_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248050": {
+            "content": "<|box_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248051": {
+            "content": "<|quad_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248052": {
+            "content": "<|quad_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248053": {
+            "content": "<|vision_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248054": {
+            "content": "<|vision_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248055": {
+            "content": "<|vision_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248056": {
+            "content": "<|image_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248057": {
+            "content": "<|video_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248058": {
+            "content": "<tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248059": {
+            "content": "</tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248060": {
+            "content": "<|fim_prefix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248061": {
+            "content": "<|fim_middle|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248062": {
+            "content": "<|fim_suffix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248063": {
+            "content": "<|fim_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248064": {
+            "content": "<|repo_name|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248065": {
+            "content": "<|file_sep|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248066": {
+            "content": "<tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248067": {
+            "content": "</tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248068": {
+            "content": "<think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248069": {
+            "content": "</think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "248070": {
+            "content": "<|audio_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248071": {
+            "content": "<|audio_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248072": {
+            "content": "<tts_pad>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248073": {
+            "content": "<tts_text_bos>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248074": {
+            "content": "<tts_text_eod>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248075": {
+            "content": "<tts_text_bos_single>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "248076": {
+            "content": "<|audio_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        }
+    },
+    "additional_special_tokens": [
+        "<|im_start|>",
+        "<|im_end|>",
+        "<|object_ref_start|>",
+        "<|object_ref_end|>",
+        "<|box_start|>",
+        "<|box_end|>",
+        "<|quad_start|>",
+        "<|quad_end|>",
+        "<|vision_start|>",
+        "<|vision_end|>",
+        "<|vision_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>"
+    ],
+    "bos_token": null,
+    "chat_template": "{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- macro render_content(content, do_vision_count, is_system_content=false) %}\n    {%- if content is string %}\n        {{- content }}\n    {%- elif content is iterable and content is not mapping %}\n        {%- for item in content %}\n            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain images.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set image_count.value = image_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Picture ' ~ image_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}\n            {%- elif 'video' in item or item.type == 'video' %}\n                {%- if is_system_content %}\n                    {{- raise_exception('System message cannot contain videos.') }}\n                {%- endif %}\n                {%- if do_vision_count %}\n                    {%- set video_count.value = video_count.value + 1 %}\n                {%- endif %}\n                {%- if add_vision_id %}\n                    {{- 'Video ' ~ video_count.value ~ ': ' }}\n                {%- endif %}\n                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}\n            {%- elif 'text' in item %}\n                {{- item.text }}\n            {%- else %}\n                {{- raise_exception('Unexpected item type in content.') }}\n            {%- endif %}\n        {%- endfor %}\n    {%- elif content is none or content is undefined %}\n        {{- '' }}\n    {%- else %}\n        {{- raise_exception('Unexpected content type.') }}\n    {%- endif %}\n{%- endmacro %}\n{%- if not messages %}\n    {{- raise_exception('No messages provided.') }}\n{%- endif %}\n{%- set reasoning_instructions = '' %}\n{%- if enable_thinking is undefined or enable_thinking is true %}\n    {%- set resolved_reasoning_effort = reasoning_effort|default('xhigh') %}\n    {%- if resolved_reasoning_effort not in ('xhigh', 'medium', 'low') %}\n        {{- raise_exception('Unexpected reasoning effort ' ~ reasoning_effort ~ '. Supported types are xhigh (default), medium, and low.') }}\n    {%- endif %}\n    {%- if resolved_reasoning_effort == 'xhigh' %}\n        {%- set reasoning_instructions = 'Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.' %}\n    {%- elif resolved_reasoning_effort == 'low' %}\n        {%- set reasoning_instructions = 'Reasoning effort is set to low. Keep your thinking brief and focused, moving directly to the conclusion without unnecessary elaboration.' %}\n    {%- endif %}\n{%- endif %}\n{%- if tools and tools is iterable and tools is not mapping %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if reasoning_instructions %}\n        {{- reasoning_instructions + '\\n\\n' }}\n    {%- endif %}\n    {{- \"# Tools\\n\\nYou have access to the following functions:\\n\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n<tool_call>\\n<function=example_function_name>\\n<parameter=example_parameter_1>\\nvalue_1\\n</parameter>\\n<parameter=example_parameter_2>\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n</parameter>\\n</function>\\n</tool_call>\\n\\n<IMPORTANT>\\nReminder:\\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n</IMPORTANT>' }}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '\\n\\n' + content }}\n        {%- endif %}\n    {%- endif %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {%- set content = render_content(messages[0].content, false, true)|trim %}\n        {%- if content %}\n            {{- '<|im_start|>system\\n' + (reasoning_instructions + '\\n\\n' if reasoning_instructions else '')  + content + '<|im_end|>\\n' }}\n        {%- elif reasoning_instructions %}\n            {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- elif reasoning_instructions %}\n        {{- '<|im_start|>system\\n' + reasoning_instructions + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" %}\n        {%- set content = render_content(message.content, false)|trim %}\n        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}\n            {%- set ns.multi_step_tool = false %}\n            {%- set ns.last_query_index = index %}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if ns.multi_step_tool %}\n    {{- raise_exception('No user query found in messages.') }}\n{%- endif %}\n{%- for message in messages %}\n    {%- set content = render_content(message.content, true)|trim %}\n    {%- if message.role == \"system\" %}\n        {%- if not loop.first %}\n            {{- raise_exception('System message must be at the beginning.') }}\n        {%- endif %}\n    {%- elif message.role == \"user\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is string %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- endif %}\n        {%- set reasoning_content = reasoning_content|trim %}\n        {%- if preserve_thinking is undefined or preserve_thinking is true or loop.index0 > ns.last_query_index %}\n            {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content + '\\n</think>\\n\\n' + content }}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if tool_call.function is defined %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {%- if loop.first %}\n                    {%- if content|trim %}\n                        {{- '\\n\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- else %}\n                        {{- '<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                    {%- endif %}\n                {%- else %}\n                    {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n                {%- endif %}\n                {%- if tool_call.arguments is defined and tool_call.arguments != '' %}\n                    {%- for args_name, args_value in tool_call.arguments|items %}\n                        {{- '<parameter=' + args_name + '>\\n' }}\n                        {%- set args_value = args_value | string if args_value is string else args_value | tojson | safe %}\n                        {{- args_value }}\n                        {{- '\\n</parameter>\\n' }}\n                    {%- endfor %}\n                {%- endif %}\n                {{- '</function>\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- content }}\n        {{- '\\n</tool_response>' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- raise_exception('Unexpected message role.') }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- else %}\n        {{- '<think>\\n' }}\n    {%- endif %}\n{%- endif %}",
+    "clean_up_tokenization_spaces": false,
+    "eos_token": "<|im_end|>",
+    "errors": "replace",
+    "model_max_length": 262144,
+    "pad_token": "<|endoftext|>",
+    "split_special_tokens": false,
+    "tokenizer_class": "Qwen2Tokenizer",
+    "unk_token": null,
+    "add_bos_token": false,
+    "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+    "extra_special_tokens": {
+        "audio_bos_token": "<|audio_start|>",
+        "audio_eos_token": "<|audio_end|>",
+        "audio_token": "<|audio_pad|>",
+        "image_token": "<|image_pad|>",
+        "video_token": "<|video_pad|>",
+        "vision_bos_token": "<|vision_start|>",
+        "vision_eos_token": "<|vision_end|>"
+    }
+}

verify_release.py

@@ -0,0 +1,152 @@
+"""Offline, bounded-memory file/index integrity check. Does not execute model code."""
+import argparse
+import hashlib
+import json
+import math
+from pathlib import Path, PurePosixPath
+import re
+import struct
+
+
+def unique(pairs):
+    result = {}
+    for key, value in pairs:
+        if key in result:
+            raise ValueError('duplicate JSON key')
+        result[key] = value
+    return result
+
+
+def read_json(path):
+    return json.loads(path.read_text(encoding='utf-8'), object_pairs_hook=unique)
+
+
+def safe_path(root, name):
+    if not isinstance(name, str) or not name or '\\' in name:
+        raise ValueError('unsafe file path')
+    p = PurePosixPath(name)
+    if p.is_absolute() or '..' in p.parts or str(p) != name:
+        raise ValueError('unsafe file path')
+    target = root.joinpath(*p.parts)
+    if any(parent.is_symlink() for parent in (target, *target.parents) if parent != root.parent):
+        raise ValueError('symlink path not allowed')
+    target.resolve().relative_to(root.resolve())
+    return target
+
+
+def file_hashes(path):
+    size = path.stat().st_size
+    sha = hashlib.sha256()
+    blob = hashlib.sha1(f'blob {size}\0'.encode())
+    with path.open('rb') as stream:
+        for chunk in iter(lambda: stream.read(8 * 1024**2), b''):
+            sha.update(chunk)
+            blob.update(chunk)
+    return size, sha.hexdigest(), blob.hexdigest()
+
+
+def tensor_header(path):
+    size = path.stat().st_size
+    with path.open('rb') as stream:
+        prefix = stream.read(8)
+        if len(prefix) != 8:
+            raise ValueError('short safetensors prefix')
+        n = struct.unpack('<Q', prefix)[0]
+        if n > 16 * 1024**2 or 8+n > size:
+            raise ValueError('invalid header boundary')
+        header = json.loads(stream.read(n), object_pairs_hook=unique)
+    widths = {'BF16': 2, 'F16': 2, 'F32': 4, 'F64': 8, 'U32': 4, 'I32': 4,
+              'U8': 1, 'I8': 1, 'U16': 2, 'I16': 2, 'U64': 8, 'I64': 8, 'BOOL': 1}
+    spans, names = [], set()
+    for name, value in header.items():
+        if name == '__metadata__':
+            continue
+        shape = value['shape']
+        if not isinstance(shape, list) or any(type(x) is not int or x < 0 for x in shape):
+            raise ValueError('invalid tensor shape')
+        start, end = value['data_offsets']
+        if type(start) is not int or type(end) is not int or not 0 <= start <= end <= size-8-n:
+            raise ValueError('invalid tensor payload boundary')
+        if end-start != math.prod(shape)*widths[value['dtype']]:
+            raise ValueError('dtype/shape byte count mismatch')
+        spans.append((start, end))
+        names.add(name)
+    cursor = 0
+    for start, end in sorted(spans):
+        if start != cursor:
+            raise ValueError('payload gap or overlap')
+        cursor = end
+    if cursor != size-8-n:
+        raise ValueError('unreferenced or truncated payload')
+    return names
+
+
+def verify(root, expected_manifest_sha256=None):
+    root = Path(root).absolute()
+    errors, checked = [], []
+    try:
+        manifest_path = safe_path(root, 'UPLOAD_MANIFEST.json')
+        _, manifest_sha, _ = file_hashes(manifest_path)
+        if expected_manifest_sha256 and expected_manifest_sha256 != manifest_sha:
+            raise ValueError('trusted manifest hash mismatch')
+        manifest = read_json(manifest_path)
+        entries = manifest['files']
+        names = [e['path'] for e in entries]
+        if len(names) != len(set(names)):
+            raise ValueError('duplicate manifest file')
+        if any(type(e['bytes']) is not int or e['bytes'] < 0 for e in entries):
+            raise ValueError('invalid manifest size')
+        if len(entries) != manifest['file_count'] or sum(e['bytes'] for e in entries) != manifest['total_bytes']:
+            raise ValueError('manifest aggregates mismatch')
+        for e in entries:
+            if e['path'] == 'UPLOAD_MANIFEST.json' or not re.fullmatch('[a-f0-9]{64}', e['sha256']):
+                raise ValueError('invalid manifest entry')
+            p = safe_path(root, e['path'])
+            if not p.is_file():
+                errors.append({'file': e['path'], 'reason': 'missing file'})
+                continue
+            size, sha, blob = file_hashes(p)
+            if size != e['bytes'] or sha != e['sha256'] or (e.get('git_blob_sha1') and blob != e['git_blob_sha1']):
+                errors.append({'file': e['path'], 'reason': 'size or full-file hash mismatch'})
+                continue
+            if p.suffix == '.json':
+                read_json(p)
+            checked.append(e['path'])
+        index_name = 'model.safetensors.index.json'
+        if index_name not in names:
+            errors.append({'file': index_name, 'reason': 'required index not in manifest'})
+        elif index_name in checked:
+            index = read_json(root/index_name)['weight_map']
+            if not isinstance(index, dict) or not index:
+                raise ValueError('invalid weight map')
+            shards = set(index.values())
+            expected_shards = {n for n in names if n.endswith('.safetensors')}
+            if shards != expected_shards:
+                errors.append({'reason': 'index/manifest shard set mismatch'})
+            actual = {}
+            for shard in sorted(shards):
+                safe_path(root, shard)
+                if shard not in checked:
+                    errors.append({'file': shard, 'reason': 'referenced shard missing or not hash-verified'})
+                    continue
+                for tensor in tensor_header(root/shard):
+                    if tensor in actual:
+                        raise ValueError('duplicate tensor across shards')
+                    actual[tensor] = shard
+            if actual != index:
+                errors.append({'reason': 'actual tensor inventory differs from weight map'})
+    except (OSError, ValueError, TypeError, KeyError, AttributeError) as exc:
+        errors.append({'reason': 'invalid or incomplete package', 'exception_type': type(exc).__name__})
+    return {'status': 'FAIL' if errors else 'PASS', 'errors': errors,
+            'fully_hashed_files': len(checked), 'manifest_pinned': bool(expected_manifest_sha256),
+            'scope': 'File integrity and tensor index only; not generation, finite-value or quality validation'}
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('snapshot', type=Path)
+    parser.add_argument('--manifest-sha256')
+    args = parser.parse_args()
+    result = verify(args.snapshot, args.manifest_sha256)
+    print(json.dumps(result, indent=2))
+    raise SystemExit(0 if result['status'] == 'PASS' else 1)

video_preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 25165824,
+        "shortest_edge": 4096
+    },
+    "patch_size": 16,
+    "temporal_patch_size": 2,
+    "merge_size": 2,
+    "image_mean": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "image_std": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "processor_class": "Qwen3VLProcessor",
+    "video_processor_type": "Qwen3VLVideoProcessor"
+}