Decision 2.0 package Decision-2.0-Sol-2B (release) from eb2d9f1df8f17fe4e7e62949675ed6a599ae9fe6-src_training_decision2

.gitattributes

@@ -35,3 +35,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 assets/DEV2.0-2B-owl-banner.png filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+assets/banner.png filter=lfs diff=lfs merge=lfs -text
+assets/index-areas.png filter=lfs diff=lfs merge=lfs -text
+assets/index-pareto.png filter=lfs diff=lfs merge=lfs -text
+assets/jevarena-types.png filter=lfs diff=lfs merge=lfs -text

ATTRIBUTIONS.md

@@ -1,6 +0,0 @@
-# Attributions
-
-- [Decision 1.0 Sol-2B](https://huggingface.co/llm-semantic-router/Decision-1.0-Sol-2B) at `ce0c018a28de16d6639b1cd203b761bf643b89e6`: direct weight origin; every weight was fine-tuned, and its own answer probabilities were the soft targets (Apache-2.0, `LICENSE`). Its own training data reach this model through its weights, among them Cosmos QA, SQuAD 2.0 answerability labels, SNLI, MultiNLI, CLINC150 and BANKING77; see its [attributions](https://huggingface.co/llm-semantic-router/Decision-1.0-Sol-2B/blob/ce0c018a28de16d6639b1cd203b761bf643b89e6/ATTRIBUTIONS.md).
-- [Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B) at `15852e8c16360a2fea060d615a32b45270f8a8fc`: upstream text backbone and tokenizer of Decision 1.0 Sol (Apache-2.0, `LICENSES/Qwen3.5-2B-LICENSE.txt`).
-- Training data of this fine-tune (not redistributed here; each keeps its own licence). Own Decision 1.0 corpora: CLINC150 (Larson et al., EMNLP-IJCNLP 2019; CC BY 3.0), BANKING77 (Casanueva et al., 2020; CC BY 4.0), MultiNLI non-fiction genres (Williams, Nangia and Bowman, NAACL 2018; OANC terms). Decision 2.0 data: GoEmotions (Demszky et al., ACL 2020; CC BY 4.0), SNLI (Bowman et al., EMNLP 2015; CC BY-SA 4.0), TyDi QA (Clark et al., TACL 2020; Apache-2.0, Wikipedia text CC BY-SA), WinoGrande (Sakaguchi et al., AAAI 2020; Apache-2.0), SQuAD 2.0 (Rajpurkar et al., ACL 2018; CC BY-SA 4.0), GSM8K (Cobbe et al., 2021; MIT), MLQE-PE (Fomicheva et al., LREC 2022; CC0 annotations, Wikipedia text CC BY-SA), 2WikiMultihopQA (Ho et al., COLING 2020; Apache-2.0, Wikipedia text CC BY-SA), HotpotQA (Yang et al., EMNLP 2018; CC BY-SA 4.0), KLUE MRC and STS (Park et al., 2021; CC BY-SA 4.0), IBM ArgQ-30k (Gretz et al., AAAI 2020; CC BY-SA 3.0), JGLUE JSQuAD and JSTS (Kurihara et al., LREC 2022; CC BY-SA 4.0), QuAC (Choi et al., EMNLP 2018; CC BY-SA 4.0, its dataset card says MIT), DRCD (Shao et al., 2018; CC BY-SA 3.0), SQAC (Gutiérrez-Fandiño et al., 2022; CC BY-SA 4.0), CMRC 2018 (Cui et al., EMNLP 2019; CC BY-SA 4.0), GermanQuAD (Möller et al., 2021; CC BY 4.0), CommonsenseQA (Talmor et al., NAACL 2019; MIT), MultiWOZ 2.2 (Zang et al., 2020; MIT), MTOP (Li et al., EACL 2021; CC BY-SA 4.0), PIAF (Keraron et al., LREC 2020; MIT), SciTail (Khot et al., AAAI 2018; Apache-2.0), MuSiQue (Trivedi et al., TACL 2022; CC BY 4.0, Wikipedia text CC BY-SA), DBpedia-14 (Zhang et al., NeurIPS 2015; CC BY-SA 3.0), Taskmaster-2 (Byrne et al., 2020; CC BY 4.0), ROPES (Lin et al., MRQA 2019; CC BY 4.0), SAF (Filighera et al., ACL 2022; CC BY 4.0), QuaRTz (Tafjord et al., EMNLP 2019; CC BY 4.0), OneStopEnglish (Vajjala and Lučić, BEA 2018; CC BY-SA 4.0), and project-generated decision tasks.
-- Evaluation data: JevArena and JevBench public 231 (panels described in `evaluation/EVALUATION.md`); mlx-diag Choice from the MASSIVE 1.1 test split (FitzGerald et al., 2022; CC BY 4.0) and Noul from the PAWS-X test split (Yang et al., EMNLP 2019).

LICENSES/Qwen3.5-2B-LICENSE.txt

@@ -1,202 +0,0 @@
-
-                                 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.

MODEL_MANIFEST.json

@@ -1,31 +1,36 @@
 {
   "base": null,
   "builder": {
-    "module_sha256": "117ca64f2f35b191226e1352cbb6b07ebf89d560a70b53c850dde90838001063",
-    "source_commit": "f85ea4e1728203aaacc7c5ecc77a15f7de9ee75d"
+    "module_sha256": "7f74839fb11762d66fa4cdc9657e7dfa8f4041bb20293dedbf69d54640fe0f0d",
+    "source_commit": "eb2d9f1df8f17fe4e7e62949675ed6a599ae9fe6"
   },
   "calibration": null,
   "card": {
+    "assets_receipt_sha256": "318aec32352e099f041351b503df7d8dd697fd225085003c3a511a990b75e2b2",
     "figures_sha256": {
-      "assets/jevarena.svg": "9bcb34999b25b561c107b329d5c320ee404e3fa4331e97570e470ad83f420fd9",
-      "assets/jevbench-public-231.svg": "b1fe51462402074e7bc485095cf7eaea0a1151cac29f4c9bf8c96f725dfaac96"
+      "assets/banner.png": "ce6c9247d0ebb9a2906fde6cc4d743d1f91ca1294d2be678135539de8c1a96c4",
+      "assets/index-areas.png": "9cc657240233939a8267191e9b0c8949cf0bfe008995721c42d213459bc4b835",
+      "assets/index-pareto.png": "994a17446d3b4a5d3416038e247256a26afc4e3ba9a77941e6b2d21998ba74f3",
+      "assets/jevarena-types.png": "34b007d8c56bacca1898d05b804db5a10c72113fe117ce3fb1e6b497f1f55316",
+      "assets/jevarena.png": "2e26f0972f90bb274b28f4d929bdfcbe58f147c257f4787019bf92000d371f7d"
     },
-    "readme_sha256": "f63aa70b5da943d14b407fc48890b858a0a26b47bb6e75156ebe70c17674dc8a"
+    "index_sha256": "47e59a9550767b07d8aa82effb3cee4d5fbde75c4abed725e6ed2e6551826a01",
+    "readme_sha256": "b215fb8248834529b18d63ac582804551e706f4b098aae4b5771dd1871d469e4"
   },
   "files_sha256": {
-    "ATTRIBUTIONS.md": "ac31e0c400a2bd1937a7dd6838db33ca0c381e2cac7cd3c85e5850ce0061282b",
     "LICENSE": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a",
-    "LICENSES/Qwen3.5-2B-LICENSE.txt": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a",
-    "NOTICE": "c2151d404543c66e1a42a528d103ffdf0b3869ce378874d688bab9b40ac6a12e",
-    "README.md": "f63aa70b5da943d14b407fc48890b858a0a26b47bb6e75156ebe70c17674dc8a",
-    "assets/jevarena.svg": "9bcb34999b25b561c107b329d5c320ee404e3fa4331e97570e470ad83f420fd9",
-    "assets/jevbench-public-231.svg": "b1fe51462402074e7bc485095cf7eaea0a1151cac29f4c9bf8c96f725dfaac96",
+    "README.md": "b215fb8248834529b18d63ac582804551e706f4b098aae4b5771dd1871d469e4",
+    "assets/banner.png": "ce6c9247d0ebb9a2906fde6cc4d743d1f91ca1294d2be678135539de8c1a96c4",
+    "assets/index-areas.png": "9cc657240233939a8267191e9b0c8949cf0bfe008995721c42d213459bc4b835",
+    "assets/index-pareto.png": "994a17446d3b4a5d3416038e247256a26afc4e3ba9a77941e6b2d21998ba74f3",
+    "assets/jevarena-types.png": "34b007d8c56bacca1898d05b804db5a10c72113fe117ce3fb1e6b497f1f55316",
+    "assets/jevarena.png": "2e26f0972f90bb274b28f4d929bdfcbe58f147c257f4787019bf92000d371f7d",
     "backbone/config.json": "071e97d8291168ba712237744c9a60e733acce554e6164322d22550dc96de19b",
     "backbone/model-00001-of-00002.safetensors": "17ba386b5f9bab7c648ef2509988165cdc341bb75c35a2fa5e8b27faece88fe0",
     "backbone/model-00002-of-00002.safetensors": "aec9a7eaadce4e7ad764614a56a8183d67ee94eb2e63969165ba129d07a0281a",
     "backbone/model.safetensors.index.json": "7607bb56aa11afe3a32bba1855e6bc3e54767b4a29c28216765f368b01da45a2",
     "chat_template.jinja": "273d8e0e683b885071fb17e08d71e5f2a5ddfb5309756181681de4f5a1822d80",
-    "config.json": "5151202ad1864c9c6cfdf28150452f6acbaef926657b1cef6d3949ecec10ae91",
+    "config.json": "60c6f61ba93b64fec2d62697cfdb8a474bd0c0299d049592615e36dfde705d03",
     "configuration_decision2.py": "b88ce8380cb0f210776f3f3ccc06219489b96502c520317790f3f4253e2a9367",
     "decision2/__init__.py": "10fea99a87686a8f5d99e9a8a91573ce1ddb7bc702f22b1fa21d3de258a4054a",
     "decision2/_vendor/__init__.py": "fe3babbdd871fa7c0cf919320a66efd1249253abe6bd7cdc5e33352dc46adeef",
@@ -40,8 +45,6 @@
     "decision2/qwen.py": "97ba565f95d5a5ac9af6632e7397c05cfe9ab47fde68b56df19aabfa2d8fb187",
     "decision_config.json": "0b6c3429ee06032739d7386659c332ed7bb62d0b96ccddfcad4fad999e22cd4f",
     "decision_head.safetensors": "33b6541bb6636677eb91a4d8e06acd4db81152b11097088840796f11d49c707a",
-    "evaluation/EVALUATION.md": "8d62877a481ed401eb8ea1660746f0e5e3ce119b0933cdddde37717abd2f0429",
-    "evaluation/manifest.json": "6dcc52e22d59020279d85a04d9c6e69776389451e98b5a4aa6a1d6fba2b407d8",
     "modeling_decision2.py": "a3f700b3d2a5deb344821802248a5ea7ea06ccedc5af1823dca4e3e813c17fbb",
     "pipeline_decision2.py": "3393faec770ddaa9045a7e41092ad17cee5310fe74187deb6f58bbe718b0c1c3",
     "tokenizer.json": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
@@ -65,9 +68,9 @@
   "licence": {
     "components": [
       {
-        "component": "DEV2.0-2B weights, decision head, package runtime, card and artwork",
+        "component": "Decision-2.0-Sol-2B weights, decision head, package runtime, card and artwork",
         "licence": "apache-2.0",
-        "source": "llm-semantic-router/DEV2.0-2B"
+        "source": "llm-semantic-router/Decision-2.0-Sol-2B"
       },
       {
         "component": "Decision 1.0 Sol-2B weights (direct weight origin, fully fine-tuned)",
@@ -94,7 +97,7 @@
     "tokenizer.json",
     "tokenizer_config.json"
   ],
-  "model_name": "DEV2.0-2B",
+  "model_name": "Decision-2.0-Sol-2B",
   "origin": {
     "relation": "finetune",
     "repo_id": "llm-semantic-router/Decision-1.0-Sol-2B",
@@ -168,7 +171,7 @@
       "5.18.0"
     ]
   },
-  "repo_id": "llm-semantic-router/DEV2.0-2B",
+  "repo_id": "llm-semantic-router/Decision-2.0-Sol-2B",
   "runtime": {
     "equivalence": "decision2/qwen.py loads this full checkpoint with the vendored training/model sources whose SHA-256 equal the scored adapter sources (checked at build time) and applies the per-item batching, BF16-backbone / FP32-head execution, raw probabilities (temperature 1; no calibration file) and answer normalization of v2.dec.infer_dec, which for this non-residual checkpoint wraps the same DecisionModel without extra readouts. The scored checkpoint (073bd1f2) stored every tensor in FP32; this package (v2.release.bf16_copy, receipt a5229ef1) stores its 186 Linear projection matrices in BF16 exactly as BF16 autocast rounds them and every other tensor bit for bit in FP32. From this revision the runtime holds the backbone's BF16-exact Linear weights in BF16, the values BF16 autocast multiplies with, instead of FP32 copies cast before every matmul; it was checked with 0 answer changes on every scored prompt and on mlx-diag (2,275). From this revision the package also ships 🤗 Transformers remote code (configuration_decision2.py, modeling_decision2.py, pipeline_decision2.py; config.json gains model_type, auto_map and custom_pipelines): AutoModel with trust_remote_code loads the package through this runtime, which now also refuses Transformers' remote-code prompt while loading the tokenizer; it was checked with 0 answer changes against the native runtime on every scored prompt and on mlx-diag (2,275). From this revision the runtime runs a request whose padded question batch would put more than 2**30 elements in a gated-delta q / k / v tensor (the FLA kernels' 32-bit offsets) as several GPU-sized batches, longest questions first; requests within that budget, every scored prompt among them, keep the single-batch path, and it was checked with 0 answer changes on every scored prompt and on mlx-diag (2,275). Checked on one GPU of the scoring node against the T = 1 predictions derived exactly from the sealed CAL698 predictions of every scored prompt (typed-final 1,600, css15 6,547, public231 231) and of the mlx-diag diagnostic (2,275) by release.sh --parity, with the scored run's persisted Triton autotune cache.",
     "requirements": {

NOTICE

@@ -1,5 +0,0 @@
-DEV2.0-2B (Decision 2.0)
-
-Original Decision 2.0 package code and model card (text and charts) are released
-under the Apache License 2.0 in LICENSE. Third-party material keeps its own
-licence and notices, listed in ATTRIBUTIONS.md.

README.md

@@ -10,50 +10,28 @@ tags:
 - safetensors
 ---
 
-# DEV2.0-2B
+![Decision-2.0-Sol-2B](assets/banner.png)
 
-DEV2.0-2B is a Decision 2.0 model. It answers structured decision questions about one input (**Choice** among named options, **Noul** yes/no checks and **Score** ratings on ordered levels) and returns a probability for every answer, all in one forward pass and without generating text. Options and rubrics are supplied with each request.
+# Decision-2.0-Sol-2B
 
-[Decision 2.0 collection](https://huggingface.co/collections/llm-semantic-router/decision-20-6ab7cf7bdfb506bf8269cb00) · [Evaluation details](evaluation/EVALUATION.md)
-
-## Highlights
-
-- **JevArena 53.44**, +7.66 over its Decision 1.0 counterpart, Decision 1.0 Sol (paired 95% CI +3.26 to +10.81).
-- Highest JevArena score of the 5 same-size models compared.
-- Runs with stock 🤗 Transformers through `AutoModel` or `pipeline("decision")` with `trust_remote_code=True`.
-
-## Model overview
+**Decision-2.0-Sol-2B** is the 2B model of [Decision 2.0](https://huggingface.co/collections/llm-semantic-router/decision-20-6ab7cf7bdfb506bf8269cb00), the decision models of [vLLM Semantic Router](https://github.com/vllm-project/semantic-router). Give it an input (text or JSON) and the questions you need answered: pick one of several options, say yes or no, or rate on a scale. It answers them all at once and returns a probability for every answer, without generating text.
 
 | | |
 | --- | --- |
-| **Model type** | Decision model: a 24-layer Qwen3.5 text backbone (gated-delta and attention layers) with a candidate-scoring head; returns probabilities, does not generate text |
-| **Base model** | [Decision-1.0-Sol-2B](https://huggingface.co/llm-semantic-router/Decision-1.0-Sol-2B) (full fine-tune; [Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B) backbone) |
-| **Parameters** | 1.88B (1,883,930,944) |
-| **Context length** | 16,384 tokens (state, questions and options together) |
-| **Decision types** | Choice (2–255 options), Noul (yes/no), Score (2–10 ordered levels) |
-| **Precision** | BF16 backbone compute, FP32 decision head |
-| **License** | [Apache-2.0](LICENSE) |
-
-## Evaluation
-
-![JevArena scores of DEV2.0-2B and same-size models](assets/jevarena.svg)
-
-![JevBench public 231 scores of DEV2.0-2B and same-size models](assets/jevbench-public-231.svg)
+| **Parameters** | 1.88B |
+| **Context length** | 16,384 tokens |
+| **Decision types** | Choice · Yes / No · Score |
+| **License** | Apache-2.0 |
 
-| Model | Parameters | JevArena ↑ | Human-labelled transfer ↑ | JevBench (public 231) ↑ |
-| --- | ---: | ---: | ---: | ---: |
-| **DEV2.0-2B** | **1.88B** | **53.44** | **52.5** | **171** |
-| Decider 2B | 1.88B | 49.50 | 42.0 | 175 |
-| This-That 1.2 | 1.88B | 46.11 | 40.5 | 147 |
-| Decision 1.0 Sol | 1.88B | 45.78 | 49.3 | 160 |
-| Bosun v3.1 1.7B | 1.74B | 42.12 | 38.0 | 151 |
+## Highlights
 
-<sub>Every model ran on the same frozen prompts with the same scorers; missing or invalid answers count as errors. JevArena combines typed-decision accuracy with the median macro-F1 of 15 human-labelled transfer tasks (shown ×100). Methods, per-task results and comparator notes: [EVALUATION.md](evaluation/EVALUATION.md).</sub>
+- **Top JevArena score of its size:** 53.4, ahead of the 4 other same-size models compared.
+- **Ahead of Decision 1.0 Sol:** +7.7 on JevArena and +1.5 on the Jev Decision Index.
+- **Speed:** a median of 22.8 ms per single-question request on a single GPU.
+- **Many questions, one pass:** Choice, Yes / No and Score questions about the same input are answered together in one forward pass, with a probability for every option.
 
 ## Quickstart
 
-### Use with 🤗 Transformers
-
 ```bash
 pip install "transformers>=5.17" torch safetensors
 ```
@@ -63,7 +41,7 @@ import json
 
 from transformers import AutoModel
 
-model = AutoModel.from_pretrained("llm-semantic-router/DEV2.0-2B", trust_remote_code=True)
+model = AutoModel.from_pretrained("llm-semantic-router/Decision-2.0-Sol-2B", trust_remote_code=True)
 result = model.system_one(
     state="The order arrived damaged yesterday. The customer has a receipt and asks for a replacement today.",
     questions={
@@ -92,36 +70,48 @@ result = model.system_one(
     },
 )
 print(json.dumps(result["answers"], indent=2))
+
+# Or as a pipeline:
+# transformers.pipeline("decision", model="llm-semantic-router/Decision-2.0-Sol-2B", trust_remote_code=True)(state=..., questions=...)
 ```
 
-`pipeline("decision", model="llm-semantic-router/DEV2.0-2B", trust_remote_code=True)` accepts `{"state": ..., "questions": {...}}` and returns the same answers. The model loads on `cuda:0`; pass `device_map="cuda:1"` to choose another GPU. Install `flash-linear-attention` and `causal-conv1d` for the kernels the evaluation used; without them Transformers runs slower reference code whose probabilities differ in the last digits. CPU inference was not verified. Tested with Transformers 5.17.0 and 5.18.0.
+## Evaluation
+
+| Model | JevArena ↑ | Human-labelled transfer ↑ | Jev Decision Index ↑ |
+| --- | ---: | ---: | ---: |
+| **Decision-2.0-Sol-2B** | **53.4** | **52.5** | **26.8** |
+| Decider 2B | 49.5 | 42.0 | — |
+| This-That 1.2 | 46.1 | 40.5 | — |
+| Decision 1.0 Sol | 45.8 | 49.3 | 25.3 |
+| Bosun v3.1 1.7B | 42.1 | 38.0 | — |
+
+### JevArena
+
+![JevArena: Decision-2.0-Sol-2B and same-size models](assets/jevarena.png)
 
-## Limitations
+![JevArena by decision type: Decision-2.0-Sol-2B and same-size models](assets/jevarena-types.png)
 
-- **Below Decision 1.0 Sol in places:** 8 of 15 human-labelled transfer tasks (largest gaps, macro-F1: Misinfo Reaction Frames 53.7 vs 63.1, Wikipedia power 45.0 vs 49.5) and non-English Choice and Noul on a multilingual diagnostic (66.5% vs 67.1%; 65.8% vs 68.0%). Every such result is listed in [EVALUATION.md](evaluation/EVALUATION.md).
-- **Score levels:** it almost never predicts the lowest of five typed Score levels (6 of 400 predictions), which Decision 1.0 Sol does predict.
-- Complete inputs above 16,384 tokens are rejected, never truncated. The evaluation panels are almost entirely English, so other languages are less well measured.
-- It decides only from the input it is given and does not retrieve missing facts; probabilities are estimates, so review consequential decisions against the evidence.
+<sub>Every model answers the same frozen prompts, scored the same way; missing or invalid answers count as errors. Human-labelled transfer is the median macro-F1 over 15 human-labelled tasks (×100).</sub>
 
-## Training data
+### Jev Decision Index
 
-Full fine-tune of Decision 1.0 Sol on 56,198 decision rows in 20 languages: a replay of our Decision 1.0 corpora and Decision 2.0 data (verifiable rules, counterfactuals, hard negatives, multi-level Score tasks and human labels from public reading-comprehension, commonsense, dialogue, classification and rating datasets). Decision 1.0 Sol's own probabilities were the only soft targets; the release is the uniform weight average of three seeds. No outputs of Jev or of any third-party decision model were used, and the training data were screened for overlap with the evaluation panels.
+![Jev Decision Index against model size](assets/index-pareto.png)
 
-**Base model licence.** Built on [Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B) (Apache-2.0; [licence text](LICENSES/Qwen3.5-2B-LICENSE.txt)); the upstream licence terms and copyright notices are kept with this repository.
+![Jev Decision Index by area: Decision-2.0-Sol-2B and Decision 1.0 Sol](assets/index-areas.png)
 
-**Attribution.** The training data include material under Creative Commons licences: CC BY 3.0: CLINC150; CC BY 4.0: BANKING77, GermanQuAD, GoEmotions, MuSiQue, QuaRTz, ROPES, SAF, Taskmaster-2; CC BY-SA 3.0: DBpedia-14, DRCD, IBM ArgQ-30k; CC BY-SA 4.0: CMRC 2018, HotpotQA, JGLUE JSQuAD, JGLUE JSTS, KLUE MRC, KLUE STS, MTOP, OneStopEnglish, QuAC, SNLI, SQAC, SQuAD 2.0; Wikipedia text under CC BY-SA: MLQE-PE, MuSiQue, TyDi QA. These datasets are not redistributed here; full credits, citations and licences are in [ATTRIBUTIONS.md](ATTRIBUTIONS.md).
+<sub>Decision 2.0: independent reproduction with the official 0.2.1 kit on the released weights; others: public board snapshot, 2026-09-28.</sub>
 
 ## License
 
-Apache-2.0 ([LICENSE](LICENSE)). Third-party notices: [NOTICE](NOTICE) and [ATTRIBUTIONS.md](ATTRIBUTIONS.md).
+Apache-2.0 ([LICENSE](LICENSE)).
 
 ## Citation
 
 ```bibtex
-@misc{dev2_0_2b_2026,
-  title        = {{DEV2.0-2B}: A Decision 2.0 Model for Structured Decisions},
+@misc{decision_2_0_sol_2b_2026,
+  title        = {{Decision-2.0-Sol-2B}: A Decision 2.0 Model for Structured Decisions},
   author       = {{vLLM Semantic Router Team}},
   year         = {2026},
-  howpublished = {\url{https://huggingface.co/llm-semantic-router/DEV2.0-2B}}
+  howpublished = {\url{https://huggingface.co/llm-semantic-router/Decision-2.0-Sol-2B}}
 }
 ```

config.json

@@ -32,7 +32,7 @@
   "manifest": "MODEL_MANIFEST.json",
   "max_input_tokens": 16384,
   "model_config": "decision_config.json",
-  "model_name": "DEV2.0-2B",
+  "model_name": "Decision-2.0-Sol-2B",
   "model_type": "decision2",
   "package_schema": "dev2-package/1",
   "runtime": {

evaluation/EVALUATION.md

@@ -1,67 +0,0 @@
-# DEV2.0-2B: evaluation
-
-## Method
-
-**JevArena** scores 1,600 typed decision items (Choice, Noul and Score; 2,000 answer slots) and 15 human-labelled transfer tasks (6,547 items) as `100 × sqrt(typed × transfer)`: *typed* is the macro accuracy over the four typed families and *transfer* the median macro-F1 over the 15 tasks. Missing, invalid and over-budget answers count as errors in every denominator. The panel's answers were available during development (post-key), so the results are same-panel comparisons, not an untouched blind test.
-
-JevBench public 231 is an independent rerun of the 231 public questions (about a third of the official Intelligence inputs), not the official JevBench score; its easy tier is at ceiling, and totals within about 10 items are not distinguishable.
-
-Every model ran through its own native inference path on the same frozen prompts and was scored by the same scorers; each row reports the parameters its native loader instantiates. Typed Brier and ECE (10 bins) measure calibration on the typed decisions. Paired intervals come from a joint bootstrap over typed groups (within family) and transfer tasks then items, 5,000 replicates. The DEV2.0-2B minus Decision 1.0 Sol JevArena difference is +7.66 (paired 95% interval [+3.26, +10.81]).
-
-**JevArena-C1** is an independent confirmation set of 2,840 human-labelled items from eight sources published after the relevant cutoffs, never used for training or development and scored once. DEV2.0-2B scored 45.70: level with Decision 1.0 Sol at 45.03 (+0.68, 95% CI −0.91 to +2.22) and significantly above Decider 2B (42.45) and This-That 1.2 (42.81).
-
-**mlx-diag** (a development diagnostic, not a release score: 2,275 prompts in seven languages, English instructions over target-language states): Choice comes from the MASSIVE 1.1 test split (CC BY 4.0) and Noul from the PAWS-X test split; the column shows mean non-English accuracy. Its Score part is built from XNLI (CC BY-NC 4.0) and is not shown. Public test splits may appear in backbone pretraining data, so this is not a sealed test.
-
-Comparators under non-commercial, research-only or unknown licences are not shown. Decision 1.0 Sol is shown from its stricter same-renderer run at this model's 16,384-token limit (JevArena 45.78; its adopted run scores 45.58). Decider 2B (Apache-2.0) and This-That 1.2 (MIT) declare their licences in model card metadata only; their repositories have no LICENSE file.
-
-## Results
-
-| Model | Loaded parameters | JevArena | Typed decisions | Human-labelled transfer | Choice / Noul / Score | JevBench public 231 (easy / standard / hard) | Typed Brier / ECE | mlx-diag non-English Choice / Noul | Invalid typed / transfer / public |
-| --- | ---: | ---: | ---: | ---: | --- | --- | --- | --- | --- |
-| DEV2.0-2B | 1,883,930,944 | 53.44 | 54.3 | 52.5 | 445/800 / 567/800 / 175/400 | 171 (48 / 66 / 57) | 0.271 / 0.103 | 66.5 / 65.8 | 0 / 4 / 0 |
-| Decider 2B | 1,881,825,088 | 49.50 | 58.3 | 42.0 | 545/800 / 497/800 / 133/400 | 175 (48 / 64 / 63) | 0.259 / 0.063 | 75.4 / 66.0 | 0 / 0 / 0 |
-| This-That 1.2 | 1,881,825,088 | 46.11 | 52.5 | 40.5 | 529/800 / 431/800 / 79/400 | 147 (48 / 64 / 35) | 0.344 / 0.260 | 76.2 / 66.5 | 0 / 169 / 37 |
-| Decision 1.0 Sol | 1,883,930,944 | 45.78 | 42.5 | 49.3 | 374/800 / 438/800 / 155/400 | 160 (48 / 66 / 46) | 0.345 / 0.249 | 67.1 / 68.0 | 0 / 4 / 0 |
-| Bosun v3.1 1.7B | 1,737,985,024 | 42.12 | 46.7 | 38.0 | 415/800 / 493/800 / 99/400 | 151 (46 / 62 / 43) | 0.312 / 0.127 | 70.0 / 70.2 | 0 / 0 / 0 |
-
-Typed decisions and human-labelled transfer are shown ×100.
-
-## Per-task results
-
-| Task | DEV2.0-2B | Decider 2B | This-That 1.2 | Decision 1.0 Sol | Bosun v3.1 1.7B |
-| --- | ---: | ---: | ---: | ---: | ---: |
-| Typed Choice (accuracy) | 55.6 | 68.1 | 66.1 | 46.8 | 51.9 |
-| Typed Noul (accuracy) | 70.9 | 62.1 | 53.9 | 54.8 | 61.6 |
-| Typed Score (accuracy) | 43.8 | 33.2 | 19.8 | 38.8 | 24.8 |
-| Conversations Gone Awry (macro-F1) | 44.1 | 35.9 | 35.2 | 44.3 | 37.9 |
-| Emotion (macro-F1) | 57.2 | 77.7 | 78.5 | 49.3 | 41.1 |
-| FLUTE figurative language (macro-F1) | 62.8 | 44.3 | 33.4 | 66.1 | 46.2 |
-| Ideological Books Corpus (macro-F1) | 40.1 | 32.2 | 27.5 | 42.7 | 42.2 |
-| Indian English dialect (macro-F1) | 28.3 | 37.7 | 28.9 | 25.8 | 25.8 |
-| Media ideology (macro-F1) | 35.1 | 31.0 | 25.6 | 30.5 | 28.3 |
-| Misinfo Reaction Frames (macro-F1) | 53.7 | 74.3 | 69.0 | 63.1 | 72.0 |
-| Persuasion (macro-F1) | 57.2 | 51.7 | 55.6 | 58.0 | 46.0 |
-| Random Acts of Pizza (macro-F1) | 58.8 | 49.4 | 51.3 | 54.3 | 38.0 |
-| Reddit humour (macro-F1) | 52.5 | 56.9 | 59.6 | 41.8 | 37.2 |
-| TalkLife empathy (macro-F1) | 31.9 | 34.8 | 40.5 | 32.0 | 35.1 |
-| TempoWiC (macro-F1) | 58.5 | 55.5 | 46.6 | 52.7 | 48.0 |
-| Character tropes (macro-F1) | 4.5 | 2.2 | 0.0 | 2.6 | 2.3 |
-| Wikipedia power (macro-F1) | 45.0 | 41.3 | 54.6 | 49.5 | 37.7 |
-| Wikipedia politeness (macro-F1) | 55.0 | 42.0 | 26.0 | 57.4 | 47.7 |
-
-## Results below Decision 1.0 Sol
-
-| Result | DEV2.0-2B | Decision 1.0 Sol |
-| --- | ---: | ---: |
-| mlx-diag non-English Choice (accuracy) | 66.5% | 67.1% |
-| mlx-diag non-English Noul (accuracy) | 65.8% | 68.0% |
-| Transfer: Misinfo Reaction Frames (macro-F1) | 53.7% | 63.1% |
-| Transfer: Wikipedia power (macro-F1) | 45.0% | 49.5% |
-| Transfer: FLUTE figurative language (macro-F1) | 62.8% | 66.1% |
-| Transfer: Ideological Books Corpus (macro-F1) | 40.1% | 42.7% |
-| Transfer: Wikipedia politeness (macro-F1) | 55.0% | 57.4% |
-| Transfer: Persuasion (macro-F1) | 57.2% | 58.0% |
-| Transfer: TalkLife empathy (macro-F1) | 31.9% | 32.0% |
-| Transfer: Conversations Gone Awry (macro-F1) | 44.1% | 44.3% |
-
-Report, panel and figure digests: [manifest.json](manifest.json).

evaluation/manifest.json

@@ -1,151 +0,0 @@
-{
-  "excluded_comparators": 0,
-  "figures_sha256": {
-    "assets/jevarena.svg": "9bcb34999b25b561c107b329d5c320ee404e3fa4331e97570e470ad83f420fd9",
-    "assets/jevbench-public-231.svg": "b1fe51462402074e7bc485095cf7eaea0a1151cac29f4c9bf8c96f725dfaac96"
-  },
-  "models": [
-    {
-      "family": "decision2",
-      "jevarena": 53.436879311778604,
-      "label": "DEV2.0-2B",
-      "licence": "package",
-      "loaded_parameters": 1883930944,
-      "mlx_diag_non_english": {
-        "choice": 0.6653439153439153,
-        "noul": 0.6583333333333333
-      },
-      "mlx_diag_sha256": "63560f76d932d0e5e106efe60525b49706019a23cd704396e6dabcc1cf0a37a5",
-      "public231_correct": 171,
-      "repo_id": null,
-      "report_loaded_parameters": 1883930944,
-      "report_sha256": "08745acf9fc753bc1d3b0335dc3e341c697a657a9e6c7c26d85eebfebf4d160c",
-      "revision": null,
-      "role": "candidate",
-      "seal_sha256": "91a3df051724344b4c4d18bddb2215a81636af98de0d74f9eb9e7f286d88beab",
-      "transfer": 0.5254514218436512,
-      "typed": 0.5434375,
-      "typed_brier": 0.2705998285734389,
-      "typed_ece_10": 0.10339657499126552
-    },
-    {
-      "family": "peer",
-      "jevarena": 49.499222791504764,
-      "label": "Decider 2B",
-      "licence": "apache-2.0",
-      "loaded_parameters": 1881825088,
-      "mlx_diag_non_english": {
-        "choice": 0.753968253968254,
-        "noul": 0.66
-      },
-      "mlx_diag_sha256": "249450b57c6d6ff15b5073bea1915da2c9db9dab241b311539fd41a7ce14d21b",
-      "public231_correct": 175,
-      "repo_id": "Mapika/decider-2b",
-      "report_loaded_parameters": 1881825088,
-      "report_sha256": "dd91e5ba628dbb5a3d2fc4092e9fd62d336c0b88f732419c7dfadd83f634e8a2",
-      "revision": "533964dae8be954c5b5e19fa4948e48408094c1e",
-      "role": "peer",
-      "seal_sha256": "1131965e01aed5b397ca29b8bb20d662ea09eae997936baaf76870e147366c24",
-      "transfer": 0.42017973109762485,
-      "typed": 0.583125,
-      "typed_brier": 0.2592733321913117,
-      "typed_ece_10": 0.06303489578140631
-    },
-    {
-      "family": "peer",
-      "jevarena": 46.11187301083141,
-      "label": "This-That 1.2",
-      "licence": "mit",
-      "loaded_parameters": 1881825088,
-      "mlx_diag_non_english": {
-        "choice": 0.7619047619047619,
-        "noul": 0.665
-      },
-      "mlx_diag_sha256": "860e8dd232f4fa5771d77aff7e1db6e4aac0481e45cef61efa8b246307f643a8",
-      "public231_correct": 147,
-      "repo_id": "flock-io/this-that-model-1.2",
-      "report_loaded_parameters": 1881825088,
-      "report_sha256": "33daa04aa94d831b630401875dfec50c16cfc63e7214be32eb9008807dd29ea3",
-      "revision": "c4d1c30b8d512d278726de439b8cc81fccc70c8f",
-      "role": "peer",
-      "seal_sha256": "4ad0a32db1a74fa27b5c4586554702eadffe422d059f3fd70ca978fe5fe4a744",
-      "transfer": 0.4050104442984842,
-      "typed": 0.525,
-      "typed_brier": 0.34352184932185165,
-      "typed_ece_10": 0.2604558593715919
-    },
-    {
-      "family": "decision1",
-      "jevarena": 45.78094715066313,
-      "label": "Decision 1.0 Sol",
-      "licence": "apache-2.0",
-      "loaded_parameters": 1883930944,
-      "mlx_diag_non_english": {
-        "choice": 0.6706349206349207,
-        "noul": 0.6799999999999999
-      },
-      "mlx_diag_sha256": "35641629ef5600b81a2efbcff7284f6e8a3af1a6544e01008af664fd7d1f1cf0",
-      "public231_correct": 160,
-      "repo_id": "llm-semantic-router/Decision-1.0-Sol-2B",
-      "report_loaded_parameters": 1883930944,
-      "report_sha256": "09892d3e6ebaa8196a0f72184c05f303c5f8d1773dcb43c7b52c85dd5d39ffbb",
-      "revision": null,
-      "role": "own-1.0",
-      "seal_sha256": "f16e93b262f61effffbf7446c06fbb7de9a77f65b9c8c005c0120ed927fb520b",
-      "transfer": 0.49278944823202014,
-      "typed": 0.42531250000000004,
-      "typed_brier": 0.3445250198063835,
-      "typed_ece_10": 0.2489012401486622
-    },
-    {
-      "family": "peer",
-      "jevarena": 42.117469209734864,
-      "label": "Bosun v3.1 1.7B",
-      "licence": "apache-2.0",
-      "loaded_parameters": 1737985024,
-      "mlx_diag_non_english": {
-        "choice": 0.6997354497354498,
-        "noul": 0.7016666666666667
-      },
-      "mlx_diag_sha256": "0b63232b46dc30f8c7265154fdb417985586aea2a2f2a8181022e21e336760e6",
-      "public231_correct": 151,
-      "repo_id": "Hanno-Labs/bosun-v3.1-1.7b",
-      "report_loaded_parameters": 1737985024,
-      "report_sha256": "7003d0ecead9db7710943cf350ade4a3096962ca816eb81dfc962f67abe9895c",
-      "revision": "1d8dc82a20e4a32ed60927a47272d6efff48eed2",
-      "role": "peer",
-      "seal_sha256": "0206edd184f45803e697795940d25c28b982ad70049ddae2a8bc0ab073710d7a",
-      "transfer": 0.3799477831610097,
-      "typed": 0.46687500000000004,
-      "typed_brier": 0.3117068390822657,
-      "typed_ece_10": 0.12704599892453267
-    }
-  ],
-  "paired_vs_own_1_0": {
-    "ci95": [
-      3.2646522505551383,
-      10.805032166921862
-    ],
-    "delta": 7.655932161115473,
-    "replicates": 5000,
-    "sha256": "bca44e5f286312f2e96ac69b590347ddf37d81899dd762f5ce48475365c824ab"
-  },
-  "panel_sha256": {
-    "gold/css15.gold.jsonl": "1cda9623032138bb7b124be0c1b0a4239c06bed7be3169264e6eb31805c19ba4",
-    "gold/public231/manifest.json": "e0e7c67701cf05f996d3b4eac09abf1bb3e6bb363cfe007089acd3644a38ed35",
-    "gold/public231/prompts.jsonl": "642d3fac1b6521fe33df72f9228e4e4e364b7be7ea277893207f97da5bc75ddd",
-    "gold/public231/targets.jsonl": "abc17b971d13807a15b3cdb43062f4cd876aad9d7314e72365724904e88b937f",
-    "gold/typed-final.gold.jsonl": "707dd28dfbab10d124d437434023729f319501542e999e536fce9b7ff7f2361e",
-    "goldfree/css15.prompts.jsonl": "7a527357e8ac3ca8da8f8663da66684d04c568a8c728261125c194294dd34af6",
-    "goldfree/public231.prompts.jsonl": "642d3fac1b6521fe33df72f9228e4e4e364b7be7ea277893207f97da5bc75ddd",
-    "goldfree/typed-final.prompts.jsonl": "e2a4a86bc978fc7497823e106533d8aa896a0307453d712f7bf99ee3174e87bd"
-  },
-  "schema": "dev2-card-evaluation/2",
-  "scope": "post-key same-panel",
-  "scorer_sha256": {
-    "benchmark/score.py": "d02a3b2bbaa08ec45928fc354532b3c3b5aef80e0a5d8e9ed6348ad6d30e2bcc",
-    "jev_arena/compare_v3.py": "bdb966ede0018ebdcd486c77f813ba722bc8d5f78ffab0dbf673631249d2513e",
-    "jev_arena/jevbench_public.py": "aec840b6497beeae8b63e9270670c22506265c086a84e889bc82f694146518cf",
-    "transfer/score.py": "cfe199a1826bb89b27c9eb746f808d74f16b46ca6585ac7b0ff7e440d44eeaca"
-  }
-}