Qwen3.6-35B-A3B packed for Splash: package, manifest, and model card

LICENSE

@@ -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.

README.md

@@ -0,0 +1,288 @@
+---
+license: apache-2.0
+library_name: splash
+pipeline_tag: text-generation
+inference: false
+base_model:
+  - mlx-community/Qwen3.6-35B-A3B-4bit
+  - incoai/Qwen3.6-35B-A3B-DFlash2
+base_model_relation: quantized
+tags:
+  - splash
+  - apple-silicon
+  - metal
+  - local-inference
+  - dflash2
+  - speculative-decoding
+  - qwen3.6
+  - moe
+  - 4-bit
+---
+
+# Qwen3.6-35B-A3B-Splash
+
+**Qwen3.6-35B-A3B, packed for Splash on Apple silicon.**
+
+[Splash](https://github.com/incoai/splash) is Inco AI's open-source inference
+engine for Apple silicon, built around the model. Each model on its roster
+gets its own kernels, its own DFlash 2 draft, and a memory plan computed for
+the machine it runs on.
+
+This package is what Splash maps into memory when you serve Qwen3.6-35B-A3B:
+the 4-bit target, its DFlash 2 draft, the vision encoder, and the tokenizer,
+laid out as fixed runtime artifacts. It is not a Transformers or MLX
+checkpoint and does not load in those frameworks.
+
+Qwen3.6-35B-A3B is the faster of the two launch models in Splash's benchmarks,
+a mixture-of-experts model with 35B total parameters and about 3B active per
+token. The other,
+[Qwen3.8-27B-Splash](https://huggingface.co/incoai/Qwen3.8-27B-Splash), is a
+dense 27B model.
+
+[Engine](https://github.com/incoai/splash) ·
+[Launch post and benchmarks](https://inco.ai/blog/splash/) ·
+[DFlash 2](https://inco.ai/blog/dflash2/)
+
+## Model details
+
+### Description
+
+- **Developed by:** Inco AI. Qwen3.6-35B-A3B is by Alibaba's Qwen team.
+- **Model type:** a Splash package: the 4-bit Qwen3.6-35B-A3B target, its DFlash 2 draft
+  model, the vision encoder, and the tokenizer, as fixed-layout runtime
+  artifacts.
+- **Languages:** as Qwen3.6-35B-A3B.
+- **License:** Apache-2.0. Every upstream component is Apache-2.0 as well:
+  Qwen3.6-35B-A3B (Alibaba), its 4-bit conversion (mlx-community), and the DFlash 2 draft
+  (Inco AI).
+- **Packaged from:** [`mlx-community/Qwen3.6-35B-A3B-4bit`](https://huggingface.co/mlx-community/Qwen3.6-35B-A3B-4bit)
+  for the target, tokenizer, and vision encoder, and
+  [`incoai/Qwen3.6-35B-A3B-DFlash2`](https://huggingface.co/incoai/Qwen3.6-35B-A3B-DFlash2) for the draft, at the
+  revisions listed under Technical specifications.
+
+### Sources
+
+- **Engine:** <https://github.com/incoai/splash>
+- **Launch post and benchmarks:** <https://inco.ai/blog/splash/>
+- **Draft model:** <https://inco.ai/blog/dflash2/>
+- **Companion package:** [Qwen3.8-27B-Splash](https://huggingface.co/incoai/Qwen3.8-27B-Splash)
+
+## Uses
+
+### Direct use
+
+Serve Qwen3.6-35B-A3B locally with Splash: to coding agents (OpenCode, Claude
+Code, Codex CLI, Hermes Agent) and to any client of the OpenAI Chat
+Completions, Responses, or Anthropic Messages APIs. Text, images, and inline
+PDFs in. Streaming, tool calls, and JSON Schema output out.
+
+### Out-of-scope use
+
+- Loading it anywhere but Splash. It is not a Transformers, MLX, or GGUF
+  checkpoint. For other engines, use the upstream checkpoints.
+- Using the draft on its own. It is not a standalone language model.
+- Anything Qwen3.6-35B-A3B's own license and usage policy exclude.
+
+## Bias, risks, and limitations
+
+- The package inherits Qwen3.6-35B-A3B's biases, knowledge cutoff, and failure modes. See
+  the upstream model card.
+- The target is the upstream 4-bit conversion. We have not published a quality
+  comparison against the bf16 checkpoint.
+- The draft proposes tokens and the target verifies every one, so speculative
+  decoding changes the speed and not the output distribution.
+- Reasoning is on by default. Reasoning tokens count toward the output limit and
+  the time to the final answer.
+- Usable context is bounded by the memory left after the weights, so a 36 GB
+  Mac holds less context and fewer concurrent requests than a 48 GB one.
+- The server binds `127.0.0.1` and has no authentication. Put a proxy in front
+  of it before exposing it to a network.
+
+## How to get started
+
+### Requirements
+
+Apple M3 or newer, macOS 26.4 or later, [Homebrew](https://brew.sh), and **36
+GB of unified memory, 48 GB or more recommended**. Splash checks available
+memory at startup and stops with a budget breakdown if the model does not fit.
+More memory means more context and more concurrent requests.
+
+### Quick start
+
+```bash
+brew install incoai/tap/splash
+splash serve --model incoai/Qwen3.6-35B-A3B-Splash
+```
+
+That is the complete setup. The first run downloads this package (20.9 GB),
+verifies it, computes a memory plan for this machine, and starts the server.
+Interrupted downloads resume. Later runs need no network.
+
+When it prints its `Ready` line, the server is up. Open:
+
+- Chat: <http://127.0.0.1:8000>
+- API: <http://127.0.0.1:8000/v1>
+
+To point a coding agent at it, run one of these in another terminal:
+
+```bash
+splash opencode    # or: splash claude / splash codex / splash hermes
+```
+
+This launches the agent you already have installed and points it at the
+serving model. Keep the server terminal open. **Ctrl+C stops the server.**
+
+### Use the API
+
+Splash serves OpenAI Chat Completions and Responses, and Anthropic Messages,
+all streaming, with tool calls, JSON Schema output, images, and inline PDFs
+(up to 20 pages and 10 MiB each).
+
+```bash
+curl http://127.0.0.1:8000/v1/chat/completions \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "incoai/Qwen3.6-35B-A3B-Splash",
+    "messages": [{"role": "user", "content": "Explain speculative decoding in one sentence."}]
+  }'
+```
+
+`model` is the package ID you served, exactly as passed to `--model`. Leave it
+out to address whatever is serving. `GET /v1/models` lists it.
+
+#### Reasoning
+
+Qwen3.6-35B-A3B reasons before it answers by default. Its reasoning is a
+switch, not a dial: on Chat Completions, `"reasoning_effort": "none"` turns it
+off, the other accepted levels (`minimal` through `max`) all render the same
+prompt, and anything else is rejected with a 400. Responses takes the same
+values in `reasoning.effort`. Anthropic Messages follows the `thinking` field
+and is off unless you send `{"type": "enabled"}` or `{"type": "adaptive"}`.
+Reasoning comes back as `reasoning_content` and is counted in
+`completion_tokens_details.reasoning_tokens`.
+
+The benchmarks below were run with reasoning on.
+
+#### Context and memory
+
+The model's native context window is 256K tokens. Splash grows its 8-bit KV cache as requests need it, within the memory left
+after the weights and the requests in flight, so the usable window depends on
+the machine and the workload. The settings are ceilings that leave room for
+other work on the same Mac:
+
+```bash
+splash serve --model incoai/Qwen3.6-35B-A3B-Splash --max-memory 28G    # cap Metal allocations
+splash serve --model incoai/Qwen3.6-35B-A3B-Splash --max-context 100K  # cap one request
+```
+
+`--max-memory` is a ceiling, not a reservation. The
+[README](https://github.com/incoai/splash#settings) lists every flag with its
+default.
+
+## Training details
+
+This package adds no training. The target weights are upstream's 4-bit
+conversion of Qwen3.6-35B-A3B. The draft was trained by Inco AI against
+Qwen3.6-35B-A3B. The [DFlash 2 post](https://inco.ai/blog/dflash2/) and the
+[draft's card](https://huggingface.co/incoai/Qwen3.6-35B-A3B-DFlash2) describe
+its training.
+
+## Evaluation
+
+### Setup
+
+Measured on an M5 Pro (16-core GPU, 48 GB unified memory) with selected
+SPEED-Bench coding prompts, over HTTP, with reasoning on, with a 1,024-token
+output limit. Single-request figures are medians, and concurrent throughput is
+averaged over four waves. Decode counts reasoning tokens. The ratio in each
+cell is against the next-fastest engine we measured.
+
+### Results
+
+| Metric | Qwen3.6-35B-A3B |
+| --- | ---: |
+| Decode · short prompt | 210 tok/s (1.7×) |
+| Prefill · 32K prompt | 2,011 tok/s (1.3×) |
+| Time to first token · 32K prompt, uncached | 17 s (1.3×) |
+| Cached time to first token · 32K replay | 123 ms (6.6×) |
+| Aggregate decode · 4 concurrent short prompts | 357 tok/s (2.0×) |
+| Aggregate decode · 4 concurrent 32K prompts | 236 tok/s (3.8×) |
+
+### Summary
+
+Splash led on every measure at every prompt length we tested. The cached
+figure replays a prompt exactly. A real agent turn also pays to process the
+tokens it appends. These are end-to-end numbers for the complete package on
+the hardware above. The [launch post](https://inco.ai/blog/splash/) has the
+full comparison against oMLX, Lily, uzu, and Ollama.
+
+## Technical specifications
+
+### Package contents
+
+```
+target/      42 files   18.2 GiB   Qwen3.6-35B-A3B, 4-bit, one packed file per layer
+draft/        7 files    0.5 GiB   DFlash 2 draft model
+vision/       1 file     0.8 GiB   bf16 vision encoder (vision/model.bin)
+tokenizer/    5 files              tokenizer and chat template
+manifest.json                      provenance, geometry, and SHA-256 of every artifact
+layout.json                        section-level map of every packed file
+```
+
+The packed weights in `target/`, `draft/`, and `vision/` are fixed-layout
+binaries that Splash maps directly from disk, with `StorageN=256` tiling and
+16 KiB section alignment. The target keeps the
+upstream conversion's mixed precision (4-bit affine weights with group size
+64, 8-bit expert routers) across its 40 hybrid layers (gated DeltaNet and
+attention), 256 experts with 8 routed per token plus a shared expert, and a
+248,320-token vocabulary. The vision encoder is the target checkpoint's own
+bf16 encoder.
+
+| Component | Source | Revision |
+| --- | --- | --- |
+| Target, tokenizer, vision | [`mlx-community/Qwen3.6-35B-A3B-4bit`](https://huggingface.co/mlx-community/Qwen3.6-35B-A3B-4bit) | `38740b847e4cb78f352aba30aa41c76e08e6eb46` |
+| Draft | [`incoai/Qwen3.6-35B-A3B-DFlash2`](https://huggingface.co/incoai/Qwen3.6-35B-A3B-DFlash2) | `8e713508f0bb02f03b5cb5cabbc8d9604f924be2` |
+
+The draft is a six-layer DFlash 2 model (hidden size 2048, 32 query heads,
+8 KV heads, 2048-token sliding-window attention, non-causal) that reads the
+target's hidden states at eight layers and proposes 7 tokens per step. The
+target verifies the block of 8 in one pass. Its projections are 4-bit affine
+(group size 64, `StorageN=256`); its norms, convolution kernels, and selector
+codebooks are bf16. The package's execution
+geometry, recorded in `manifest.json`, batches up to four sequences per decode
+step, prefills 2,048 tokens per step, and allocates KV cache in 32-token blocks.
+
+The chat template is upstream's with one change: a system message that
+appears after the first turn is rendered in place instead of being rejected,
+which is what coding agents that inject instructions mid-conversation need.
+
+### Integrity
+
+`manifest.json` records the upstream revisions above, the execution geometry,
+the converter's hashes, and the size and SHA-256 of every artifact. Splash pins
+an immutable commit of this repository plus the manifest's SHA-256, checks
+every artifact's SHA-256 before publishing the package into its local model
+directory, and re-checks sizes and alignment on every start.
+
+### Compute infrastructure
+
+- **Hardware:** Apple silicon, M3 or newer, with 36 GB of unified memory or
+  more.
+- **Software:** Splash on macOS 26.4 or later. The kernels are Metal, compiled
+  for this model's shapes and shipped precompiled.
+
+## Citation
+
+```bibtex
+@misc{inco2026splash,
+  title  = {{Splash: A Local Engine Built Around the Model}},
+  author = {{Inco AI}},
+  year   = {2026},
+  month  = {September},
+  url    = {https://inco.ai/blog/splash/}
+}
+```
+
+## Contact
+
+<contact@inco.ai>, or an issue on <https://github.com/incoai/splash>.

manifest.json

@@ -0,0 +1,412 @@
+{
+  "artifact_set_sha256": "0c465f96066fe62f185273a5c58f6de0f493d9bc64b0e43f77d1229eead2440a",
+  "artifacts": [
+    {
+      "path": "draft/layer-0.bin",
+      "sha256": "6e49b64d9448bf9965826d6c92badcd40c3321a7b86066c25b7c30852cc4c5b9",
+      "size": 34324480
+    },
+    {
+      "path": "draft/layer-1.bin",
+      "sha256": "755ef7fcf8107c3a2cd5bd24e96ad726aa06174f72d628413e32fa7b7d8ffdda",
+      "size": 34324480
+    },
+    {
+      "path": "draft/layer-2.bin",
+      "sha256": "126cedd11a4b13febc299df0101a096dd6f6761e86bc0ac5db1dc6f2f78c31c1",
+      "size": 34324480
+    },
+    {
+      "path": "draft/layer-3.bin",
+      "sha256": "d96490a76307408fc2d2e42d86481f3061b04a2289193d2e8face84531b68f9e",
+      "size": 34324480
+    },
+    {
+      "path": "draft/layer-4.bin",
+      "sha256": "92103acb13eb656fc46220f07bdc4f1da0233fed8b27ab5f2877544f1147bd93",
+      "size": 34324480
+    },
+    {
+      "path": "draft/layer-5.bin",
+      "sha256": "2de274812de6d8cdf8bdfb2b6edd43da1b7b4b82e5ffe7316e36d6d50404600d",
+      "size": 34324480
+    },
+    {
+      "path": "draft/model.bin",
+      "sha256": "532f9ca34fd57b0d042f79232bea70770c28e23d69eeb2293a46e9640959926f",
+      "size": 273498112
+    },
+    {
+      "path": "layout.json",
+      "sha256": "2e67217507b9b13c613a4ea6deb2cb5bc937547a520bf79a1a4a900db62860d7",
+      "size": 255855
+    },
+    {
+      "path": "target/embedding.bin",
+      "sha256": "a7f9ff6c44ef0bf5486d1e52c6fa0f19634fb76d7fdcac901990ed1bbd57539b",
+      "size": 286081024
+    },
+    {
+      "path": "target/head.bin",
+      "sha256": "326d8301942438f3cb5c1f3f1ae51853ab72ad50ac67cecaba034fb10e88ca53",
+      "size": 286097408
+    },
+    {
+      "path": "target/layer-0.bin",
+      "sha256": "f48116252d484df38ac0b72b957bbf0dc343a59f8b3d570834ca06f00e36e8cb",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-1.bin",
+      "sha256": "c872c1a91630b620ab8ec787fba8f68318af6605ae605bb27eac782613882f2b",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-10.bin",
+      "sha256": "0499eaf879d9ac8859c1abe3c25950ee0b6b3084d44e8040fb7e6ba6208197e4",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-11.bin",
+      "sha256": "866ab4178775804d39169ec30132f010c9d431a89ef336249a127df411e2e7d1",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-12.bin",
+      "sha256": "f10c319788e01ce6f8afdbf825707ed9c58bbb9e00e70a1e203f31617614fa25",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-13.bin",
+      "sha256": "6ce396e63dbabed82b69429eacfca5e0ed6c8d8891d0cc860ad028bc74a85eb8",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-14.bin",
+      "sha256": "6c26a5f399a66523a61c14da22268ab0ba3fc28215c30cc108ca09e8928036a0",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-15.bin",
+      "sha256": "2bcef296c32a8b12dd2c896b3dcef475b7bfab6c9d89502dc66e13dbc67ca7cc",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-16.bin",
+      "sha256": "8bf4d6dea481856f385589bcd808a52abe710b19b48b86121da13b29e9788f44",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-17.bin",
+      "sha256": "e1e4529d23161aaf24dcc0914dc1c03fa7e3c7d3d4411222f563a0e41b139e09",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-18.bin",
+      "sha256": "48c051aa2d2ede66cf45edf94abc49d06a5a2a72dda79339dfd41c9bcbe742ff",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-19.bin",
+      "sha256": "677f3bfd2162f6184a9cc14b96dbc8ead8dce285e2a80e6b52386397a0fe2c68",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-2.bin",
+      "sha256": "948d5e57d2ab8fd0781980d90cdfb1ed1a3133c7f139b4df0f1ab38c05f0d1c2",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-20.bin",
+      "sha256": "962dd25046d60c5b9e1eeccd24b8984b24ee22f9fa5ac19f82db45e883d795c5",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-21.bin",
+      "sha256": "991f2146d38887cc3eede8a028a995a6fded77e31d3d06a5c22afa2e1294ae03",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-22.bin",
+      "sha256": "dc59b50c77178f7f80915394ad4b3213fc85e71fff17e763f37181f406aa8f02",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-23.bin",
+      "sha256": "20e3eb6d773c5491337bb219441ac10af1ce055f4731f687703862205e3d9bd0",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-24.bin",
+      "sha256": "704ecb6ec7a5c3862f581669806deb1be7d94b20792e87fc076905c5fbee00c9",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-25.bin",
+      "sha256": "da26a4ffd02dc2acafd7ed755f74faa2e51b7a24aae49751c1dc4de34707872f",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-26.bin",
+      "sha256": "09997fe9dfb6a20a9b4eb75c081f66eca11c44a1d57cc2154e734574136ed216",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-27.bin",
+      "sha256": "76d5a2dfbf1d9ad1cda0efa1627751cc3f774e849eee562ebe52cfa12e1a856a",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-28.bin",
+      "sha256": "2fd221c193924fbb24b8255c24f8bbc771a4a9feb2520eb33befcf1e00ff29b5",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-29.bin",
+      "sha256": "5f28c541ee7141e6b4272a94425ee15a1de2c8104941556c855aa5d795828fe5",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-3.bin",
+      "sha256": "05c156052a20a1baa3b78c04fab330f15e64d53f0b34a97e88fa53f74e6b2db0",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-30.bin",
+      "sha256": "c78d51b8e802dee1e4d845d553acd62db04505cb558d160a6dc18f7d2db902a9",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-31.bin",
+      "sha256": "79716bdda3250501a042b25df1cecdc92570f3527632f5b3033aa78e5dafc929",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-32.bin",
+      "sha256": "2c106e5b430917d056f43bb45a3154d285934f9101732a16734496e5bfbe0bb8",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-33.bin",
+      "sha256": "34a97f512c4bd302af0d89368d886a24337f253297068a6b7d7c8cdd6bd62252",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-34.bin",
+      "sha256": "85ab41f1975bf8fc42b6d5d5e727f64eab5aae1d24c0afea2726cee5258b9097",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-35.bin",
+      "sha256": "7887d3f5e2478a270827149c7c85031ff6b5d4324b891d1880eb15c196b73369",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-36.bin",
+      "sha256": "3086e83787f03b1df90984bb986197ac325385a51f88de5932e54d30d15b0d17",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-37.bin",
+      "sha256": "d952e0e17c4f137c4f70f8acc203cf999f855668eb4b7ebf045163390f2d4353",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-38.bin",
+      "sha256": "0ae30a9dd447e61d5d4f89d650fd23c5abe4bc76cab5f7d205e5fc4738613805",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-39.bin",
+      "sha256": "70be79b973be193808a23efdb4337c176ec95c35a1c0037e021122e07f846c32",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-4.bin",
+      "sha256": "d95f30e81b433b0636285d84b120db1f72bad5f0e22bd62f84352d67b49a727d",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-5.bin",
+      "sha256": "a59c1be3c30b2c12702a108472f63d01dfe32be156c43f74d2ced47fa12b1b07",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-6.bin",
+      "sha256": "bbe39e033c15f9e7a8d2eb900614450a63854e73d6dc2187304cbb45b65cfc2d",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-7.bin",
+      "sha256": "b6544739de16a0ccdedb444aa477c250e18c5a3a0ff35551d965a513ee3f5779",
+      "size": 471285760
+    },
+    {
+      "path": "target/layer-8.bin",
+      "sha256": "6da1022800157028d974dbc802f5efe4083791b652d925c3d662c56c4652ccff",
+      "size": 475201536
+    },
+    {
+      "path": "target/layer-9.bin",
+      "sha256": "3bf6a860a4f336bb24c4c8e6d491601a07d4c5fdae4d1c70f79d476b22622f9a",
+      "size": 475201536
+    },
+    {
+      "path": "tokenizer/chat_template.jinja",
+      "sha256": "2f9e9d2156f90806f09464c24681486adcc158dbe8e8030cd55f55edff0eaa66",
+      "size": 7755
+    },
+    {
+      "path": "tokenizer/config.json",
+      "sha256": "a822a9e48b0aafbe3144ec37d4fb067e178ed96615ce6e4420b3149893cc5767",
+      "size": 23591
+    },
+    {
+      "path": "tokenizer/tokenizer.json",
+      "sha256": "87a7830d63fcf43bf241c3c5242e96e62dd3fdc29224ca26fed8ea333db72de4",
+      "size": 19989343
+    },
+    {
+      "path": "tokenizer/tokenizer_config.json",
+      "sha256": "e98f1901ac6f0adff67b1d540bfa0c36ac1a0cf59eb72ed78146ef89aafa1182",
+      "size": 1139
+    },
+    {
+      "path": "tokenizer/vocab.json",
+      "sha256": "ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003",
+      "size": 6722759
+    },
+    {
+      "path": "vision/model.bin",
+      "sha256": "20ba816b644c7221d33d1b188612d096dae3e96600be2dc3d4bd1600c0d30d00",
+      "size": 901939200
+    }
+  ],
+  "converter": {
+    "adapter_sha256": "d8a306e1a428eb9c762b994a0b7c99f15aa8c6f70309f20445ce741ef7ae8636",
+    "mlx_version": "0.32.2",
+    "source_git_commit": "d441abae04c3560d250eaac3affb81a38771f60f",
+    "source_path": "install/convert.py",
+    "source_sha256": "dd4f4b60549195835ca037a0ad01586e19ecd9390c8200a6e6f9a912f066f2aa"
+  },
+  "draft": {
+    "architecture": "DFlash2DraftModel",
+    "block_size": 8,
+    "dynamic_conv_group_size": 16,
+    "dynamic_conv_kernel_size": 2,
+    "hidden_size": 2048,
+    "intermediate_size": 6144,
+    "is_causal": false,
+    "layers": 6,
+    "selector_rank": 256,
+    "selector_top_k": 16,
+    "sliding_window": 2048,
+    "target_capture_layers": [
+      1,
+      6,
+      11,
+      16,
+      22,
+      27,
+      32,
+      37
+    ]
+  },
+  "execution_geometry": {
+    "allocation_extent_target_bytes": 134217728,
+    "attention_history_group": 128,
+    "draft_proposal_tokens": 7,
+    "draft_query_rows": 8,
+    "draft_sliding_window": 2048,
+    "maximum_batch_width": 4,
+    "prefill_token_budget": 2048,
+    "target_kv_block_tokens": 32,
+    "target_verify_rows": 8
+  },
+  "format": {
+    "draft_layer_magic": "MDFD0004",
+    "name": "splash-packed-q4-moe",
+    "q4_bits": 4,
+    "q8_bits": 8,
+    "quant_group_size": 64,
+    "section_alignment_bytes": 16384,
+    "storage_n": 256,
+    "target_layer_magic": "MDFM0001",
+    "vision_magic": "MDFV0001"
+  },
+  "model": "Qwen3.6-35B-A3B",
+  "schema_version": 4,
+  "target": {
+    "architecture": "qwen3_5_moe",
+    "attention_packed_width": 9216,
+    "experts": 256,
+    "experts_per_token": 8,
+    "gdn_actual_width": 12352,
+    "gdn_packed_width": 12544,
+    "hidden_size": 2048,
+    "layer_types": [
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention",
+      "gdn",
+      "gdn",
+      "gdn",
+      "attention"
+    ],
+    "layers": 40,
+    "moe_intermediate_size": 512,
+    "shared_expert_intermediate_size": 512,
+    "vocabulary_size": 248320
+  },
+  "upstream": {
+    "draft": {
+      "repo_id": "incoai/Qwen3.6-35B-A3B-DFlash2",
+      "revision": "8e713508f0bb02f03b5cb5cabbc8d9604f924be2"
+    },
+    "target": {
+      "repo_id": "mlx-community/Qwen3.6-35B-A3B-4bit",
+      "revision": "38740b847e4cb78f352aba30aa41c76e08e6eb46"
+    },
+    "tokenizer": {
+      "repo_id": "mlx-community/Qwen3.6-35B-A3B-4bit",
+      "revision": "38740b847e4cb78f352aba30aa41c76e08e6eb46"
+    },
+    "vision": {
+      "repo_id": "mlx-community/Qwen3.6-35B-A3B-4bit",
+      "revision": "38740b847e4cb78f352aba30aa41c76e08e6eb46"
+    }
+  }
+}