Release Qev-2B decision model and distillation recipe

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* 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/tokenizer.json filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,203 @@
+
+                                 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 Qiqian Fu and Qev contributors
+   Portions Copyright 2026 Jared Palmer
+
+   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 +1,36 @@
+Qev
+Copyright 2026 Qiqian Fu and Qev contributors
+
+Qev code, documentation, original illustrations, examples, generated prediction
+results, and published adaptation weights are licensed under Apache-2.0.
+
+This project adapts conventions and implementation ideas from Kev:
+  https://github.com/jaredpalmer/kev
+  Copyright 2026 Jared Palmer
+  Apache License, Version 2.0
+
+Adaptations include Qwen delimiters, text rendering, LoRA target selection, and
+inference cache forking. Modified files carry Qev adaptation notices. The original
+Kev license and copyright notice are retained in licenses/Kev-Apache-2.0.txt.
+
+Qev uses Qwen3.5 models and tokenizers:
+  https://huggingface.co/Qwen/Qwen3.5-9B-Base
+  https://huggingface.co/Qwen/Qwen3.5-2B-Base
+  Copyright 2026 Alibaba Cloud
+  Apache License, Version 2.0
+
+The public weight package includes the Qwen tokenizer and Qev adaptation weights.
+Qwen base weights are downloaded separately. The original Qwen license is in
+licenses/Qwen3.5-Apache-2.0.txt.
+
+The optional JevBench dataset preparer downloads public evaluation tasks by
+Florian Standhartinger and contributors. It includes their MIT license,
+third-party notices, and per-question provenance with the prepared data.
+
+See THIRD_PARTY_NOTICES.md for licensing scope and external dependencies.
+Third-party datasets and services retain their own terms.
+
+Crafter visuals in the project gameplay recording:
+  https://github.com/danijar/crafter
+  Copyright (c) 2021 Danijar Hafner
+  MIT License; see licenses/Crafter-MIT.txt.

README.md

@@ -0,0 +1,58 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen3.5-2B-Base
+base_model_relation: adapter
+language:
+  - en
+  - zh
+tags:
+  - qev
+  - lora
+  - knowledge-distillation
+  - decision-model
+---
+
+# Qev-2B
+
+Qev-2B is the compact member of the Qev decision-model family. It learns option probabilities and representation responses from Qev-9B, and supports Choice, Noul and Score with the same request format.
+
+[Model weights](https://huggingface.co/AustinFu/Qev-2B) · [Source and installation](https://github.com/QiqianFu/Qev#installation) · [中文说明](https://github.com/QiqianFu/Qev/blob/main/README.zh-CN.md)
+
+| Property | Qev-2B |
+|---|---|
+| Base | Qwen3.5-2B-Base |
+| Adapter | LoRA rank 64, alpha 128 |
+| Decision head | 256 dimensions, two Transformer layers |
+| Candidate interaction | Full sibling interaction in the final full-attention layer |
+| Training | Teacher-probability cross entropy, followed by replay and representation-response distillation |
+| Final continuation | 800 steps; seed 17; response weight 0.1 |
+| Computation | BF16 backbone, FP32 decision head |
+| Inference package | Approximately 284 MiB; base weights downloaded separately |
+
+```python
+from qev import Qev
+
+model = Qev.from_pretrained("AustinFu/Qev-2B", device="cuda")
+```
+
+Install Qev from the source repository using Python 3.12 and a hardware-compatible PyTorch 2.8.0 build. The loader downloads this adaptation package and the compatible Qwen base separately. The teacher is not needed for inference. Use `revision="v0.1.0"` to select the first published release.
+
+To fine-tune on your own data, use `configs/qev-2b-finetune.json` with `python -m qev.train --init-checkpoint AustinFu/Qev-2B`. To train with a teacher, follow the [distillation guide](https://github.com/QiqianFu/Qev/blob/main/docs/distillation.md) or [中文指南](https://github.com/QiqianFu/Qev/blob/main/docs/distillation.zh-CN.md).
+
+| Benchmark | Qwen3.5-2B-Base | Qev-2B |
+|---|---:|---:|
+| Decision development · clean | 65.43 | 85.36 |
+| Transfer development · clean | 65.09 | 77.29 |
+| MMLU-Pro · 1,000 | 31.20 | 38.70 |
+| SemIf · 144 handwritten | 63.89 | 82.64 |
+| scienthoon · 873 | 53.84 | 71.94 |
+| WANLI · 256 | 50.39 | 67.58 |
+| JevBench public · 231 | 63.20 | 74.46 |
+
+Accuracy (%). Public JevBench accuracy is 172/231. These are the selected checkpoint's recorded results. The native base uses its language-model head and zero-shot prompts. The comparison therefore includes architecture, training and readout differences; it does not isolate the contribution of response distillation. [Full results](https://github.com/QiqianFu/Qev/blob/main/docs/evaluation.md).
+
+The package includes the LoRA adapter, decision head, interaction gate, tokenizer, model configuration, both training configurations, recorded evaluation results and license files. The three trained tensor files preserve the selected checkpoint exactly; optimizer state and base weights are excluded. The complete research training corpus is not distributed.
+
+Qev supports research and development of routing, rule judgments and rubric ratings over explicit options. Evaluate the model on your application's inputs; probability calibration and production reliability have not been established.
+
+Qev adaptation weights use Apache-2.0. Source datasets and the Qwen base retain their own terms. [License and attribution](THIRD_PARTY_NOTICES.md).

THIRD_PARTY_NOTICES.md

@@ -0,0 +1,38 @@
+# Licensing and attribution
+
+## Qev materials
+
+Copyright 2026 Qiqian Fu and Qev contributors.
+
+Qev's code, documentation, original SVG illustrations, example requests, and generated prediction results are provided under the [Apache License 2.0](LICENSE). The Qev adaptation weights published at [AustinFu/Qev-9B](https://huggingface.co/AustinFu/Qev-9B), including its LoRA adapter, decision head, and interaction gate, use the same license. Contributions are accepted under these terms unless explicitly stated otherwise.
+
+Third-party materials retain the terms listed below. The Qev license does not relicense training datasets, external services, or third-party recordings.
+
+## Included and adapted materials
+
+| Material | Source and copyright | License and scope |
+|---|---|---|
+| Kev conventions and adapted implementation | [Jared Palmer's Kev](https://github.com/jaredpalmer/kev), Copyright 2026 Jared Palmer | [Apache-2.0, original notice retained](licenses/Kev-Apache-2.0.txt). Qev adapts delimiter and rendering conventions, LoRA targets, and cache forking. Modified source files identify the Qev adaptation. |
+| Qwen models and tokenizers | [Qwen3.5-9B-Base](https://huggingface.co/Qwen/Qwen3.5-9B-Base) and [Qwen3.5-2B-Base](https://huggingface.co/Qwen/Qwen3.5-2B-Base), Copyright 2026 Alibaba Cloud | [Apache-2.0](licenses/Qwen3.5-Apache-2.0.txt). Tokenizers accompany the corresponding Qev checkpoints; base weights are downloaded separately. |
+| JevBench public evaluation tasks | [JevBench](https://github.com/fstandhartinger/jevbench), Copyright 2026 Florian Standhartinger and contributors | [MIT](licenses/JevBench-MIT.txt), with [upstream third-party notices](licenses/JevBench-THIRD-PARTY.md). Tasks are downloaded by the optional preparation script. The original license, third-party notice, and per-question provenance accompany the prepared dataset. |
+| Crafter visuals in the gameplay recording | [Crafter](https://github.com/danijar/crafter), Copyright 2021 Danijar Hafner | [MIT](licenses/Crafter-MIT.txt). The gameplay recording contains the environment's visual assets; the recording and Qev decision overlay were produced for this project. |
+
+The reviewed Kev version is [557598f](https://github.com/jaredpalmer/kev/tree/557598fced1dada75dfbf36ed144dce309ac6ceb); the evaluation data use [JevBench v1.4.2](https://github.com/fstandhartinger/jevbench/tree/1df665e3956d7aab7fa0208ff6c4f2d8557f9f90). The 231 downloaded public tasks carry MIT in their individual provenance records; upstream notices describe the scope of other JevBench materials and evaluated services.
+
+The README layout takes inspiration from [JevAny](https://github.com/weitianxin/JevAny). Qev's SVG illustrations and chart code are original. The GIF and MP4 demos are this project's recordings of 9B research models. The Crafter research integration used JevAny's environment adapters; this source package does not vendor those adapters or JevAny's bundled recordings.
+
+## Dependencies installed separately
+
+Dependencies are installed from their own distributions and retain their licenses:
+
+- [PyTorch](https://github.com/pytorch/pytorch/blob/main/LICENSE): BSD-3-Clause, plus its bundled third-party notices.
+- [Transformers](https://github.com/huggingface/transformers/blob/main/LICENSE), [PEFT](https://github.com/huggingface/peft/blob/main/LICENSE), [Accelerate](https://github.com/huggingface/accelerate/blob/main/LICENSE), [Safetensors](https://github.com/huggingface/safetensors/blob/main/LICENSE), [Hugging Face Hub](https://github.com/huggingface/huggingface_hub/blob/main/LICENSE), and optional [Apache Arrow](https://github.com/apache/arrow/blob/main/LICENSE.txt): Apache-2.0.
+- Development tools [pytest](https://github.com/pytest-dev/pytest/blob/main/LICENSE) and [Matplotlib](https://matplotlib.org/stable/project/license.html) retain their MIT and Matplotlib/PSF-based license terms, respectively.
+
+Their source code and dependency distributions are not vendored in Qev's Python package. This list identifies direct dependencies; their own packages include notices for transitive components.
+
+## Training data
+
+The complete research training mixture is not distributed here. Public weight availability does not make the training data available under Apache-2.0. Any future dataset release needs its own source-specific permissions and attribution. The Wikipedia-derived training subset, for example, carries CC BY-SA attribution where applicable.
+
+The small JSONL requests in `examples/` were written for Qev and are covered by its Apache-2.0 license. The benchmark result tables and prediction files contain Qev's recorded outputs, not the original benchmark question text.

adapter/adapter_config.json

@@ -0,0 +1,56 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "kasa_config": null,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "monteclora_config": null,
+  "peft_type": "LORA",
+  "peft_version": "0.21.0",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": "b1485b2fa6dfa1287294f269f5fb618e03d52d7c",
+  "target_modules": [
+    "in_proj_z",
+    "in_proj_a",
+    "in_proj_b",
+    "up_proj",
+    "q_proj",
+    "o_proj",
+    "out_proj",
+    "v_proj",
+    "in_proj_qkv",
+    "gate_proj",
+    "down_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "FEATURE_EXTRACTION",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false,
+  "velora_config": null
+}

adapter/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6d70aa4065a631b7b719d2ce53b4c8e763efa8bdc5f56bcc8bef90afe97d7d2
+size 269155824

benchmarks.json

@@ -0,0 +1,352 @@
+{
+  "date": "2026-09-28",
+  "metric": "argmax_accuracy",
+  "models": {
+    "jev": {
+      "description": "Hosted service; JevBench tested locally at Jev 1.13.0, other suites from Kev authors"
+    },
+    "kev_9b": {
+      "revision": "2629c06a",
+      "code_revision": "557598fced1dada75dfbf36ed144dce309ac6ceb",
+      "dtype": "fp32",
+      "temperature": 1.0,
+      "source": "JevBench locally rerun; other suites from author reports"
+    },
+    "qev_9b": {
+      "base": "Qwen/Qwen3.5-9B-Base",
+      "base_revision": "68c46c4b3498877f3ef123c856ecfde50c39f404",
+      "research_name": "BranchKev",
+      "run": "c21-science-wk-late1783-9b-4gpu-r64-late50x3-s17",
+      "step": 2327,
+      "dtype": "bf16",
+      "execution": "reference",
+      "seed": 17,
+      "temperature": 1.0
+    },
+    "qwen35_9b_base": {
+      "description": "Frozen native LM head; zero-shot candidate code probabilities",
+      "dtype": "bf16"
+    },
+    "qev_2b": {
+      "base": "Qwen/Qwen3.5-2B-Base",
+      "base_revision": "b1485b2fa6dfa1287294f269f5fb618e03d52d7c",
+      "dtype": "bf16",
+      "description": "Probability and representation-response distillation from Qev-9B; no candidate preview",
+      "seed": 17,
+      "continuation_steps": 800
+    },
+    "qwen35_2b_base": {
+      "description": "Frozen native LM head; zero-shot candidate code probabilities",
+      "dtype": "bf16",
+      "base": "Qwen/Qwen3.5-2B-Base",
+      "base_revision": "b1485b2fa6dfa1287294f269f5fb618e03d52d7c"
+    }
+  },
+  "results": {
+    "decision_dev_all": {
+      "total": 1468,
+      "models": {
+        "jev": {
+          "correct": 1221,
+          "accuracy": 0.8317438692098093
+        },
+        "kev_9b": {
+          "correct": 1289,
+          "accuracy": 0.8780653950953679
+        },
+        "qev_9b": {
+          "correct": 1296,
+          "accuracy": 0.8828337874659401
+        },
+        "qwen35_9b_base": {
+          "correct": 1112,
+          "accuracy": 0.7574931880108992
+        },
+        "qev_2b": {
+          "correct": 1266,
+          "accuracy": 0.8623978201634878
+        },
+        "qwen35_2b_base": {
+          "correct": 927,
+          "accuracy": 0.6314713896457765
+        }
+      }
+    },
+    "decision_dev_clean": {
+      "total": 1264,
+      "models": {
+        "jev": {
+          "correct": 1068,
+          "accuracy": 0.8449367088607594
+        },
+        "kev_9b": {
+          "correct": 1102,
+          "accuracy": 0.8718354430379747
+        },
+        "qev_9b": {
+          "correct": 1105,
+          "accuracy": 0.8742088607594937
+        },
+        "qwen35_9b_base": {
+          "correct": 982,
+          "accuracy": 0.7768987341772152
+        },
+        "qev_2b": {
+          "correct": 1079,
+          "accuracy": 0.8536392405063291
+        },
+        "qwen35_2b_base": {
+          "correct": 827,
+          "accuracy": 0.6542721518987342
+        }
+      }
+    },
+    "transfer_dev_all": {
+      "total": 764,
+      "models": {
+        "jev": {
+          "correct": 647,
+          "accuracy": 0.8468586387434555
+        },
+        "kev_9b": {
+          "correct": 620,
+          "accuracy": 0.8115183246073299
+        },
+        "qev_9b": {
+          "correct": 630,
+          "accuracy": 0.824607329842932
+        },
+        "qwen35_9b_base": {
+          "correct": 561,
+          "accuracy": 0.7342931937172775
+        },
+        "qev_2b": {
+          "correct": 584,
+          "accuracy": 0.7643979057591623
+        },
+        "qwen35_2b_base": {
+          "correct": 493,
+          "accuracy": 0.6452879581151832
+        }
+      }
+    },
+    "transfer_dev_clean": {
+      "total": 656,
+      "models": {
+        "jev": {
+          "correct": 562,
+          "accuracy": 0.8567073170731707
+        },
+        "kev_9b": {
+          "correct": 539,
+          "accuracy": 0.8216463414634146
+        },
+        "qev_9b": {
+          "correct": 551,
+          "accuracy": 0.8399390243902439
+        },
+        "qwen35_9b_base": {
+          "correct": 488,
+          "accuracy": 0.7439024390243902
+        },
+        "qev_2b": {
+          "correct": 507,
+          "accuracy": 0.7728658536585366
+        },
+        "qwen35_2b_base": {
+          "correct": 427,
+          "accuracy": 0.6509146341463414
+        }
+      }
+    },
+    "mmlupro": {
+      "total": 1000,
+      "models": {
+        "jev": {
+          "correct": 835,
+          "accuracy": 0.835
+        },
+        "kev_9b": {
+          "correct": 511,
+          "accuracy": 0.511
+        },
+        "qev_9b": {
+          "correct": 546,
+          "accuracy": 0.546
+        },
+        "qwen35_9b_base": {
+          "correct": 504,
+          "accuracy": 0.504
+        },
+        "qev_2b": {
+          "correct": 387,
+          "accuracy": 0.387
+        },
+        "qwen35_2b_base": {
+          "correct": 312,
+          "accuracy": 0.312
+        }
+      }
+    },
+    "semif_handwritten": {
+      "total": 144,
+      "models": {
+        "jev": {
+          "correct": 139,
+          "accuracy": 0.9652777777777778
+        },
+        "kev_9b": {
+          "correct": 131,
+          "accuracy": 0.9097222222222222
+        },
+        "qev_9b": {
+          "correct": 135,
+          "accuracy": 0.9375
+        },
+        "qwen35_9b_base": {
+          "correct": 130,
+          "accuracy": 0.9027777777777778
+        },
+        "qev_2b": {
+          "correct": 119,
+          "accuracy": 0.8263888888888888
+        },
+        "qwen35_2b_base": {
+          "correct": 92,
+          "accuracy": 0.6388888888888888
+        }
+      }
+    },
+    "scienthoon": {
+      "total": 873,
+      "models": {
+        "jev": {
+          "correct": 657,
+          "accuracy": 0.7525773195876289
+        },
+        "kev_9b": {
+          "correct": 659,
+          "accuracy": 0.7548682703321878
+        },
+        "qev_9b": {
+          "correct": 631,
+          "accuracy": 0.722794959908362
+        },
+        "qwen35_9b_base": {
+          "correct": 601,
+          "accuracy": 0.6884306987399771
+        },
+        "qev_2b": {
+          "correct": 628,
+          "accuracy": 0.7193585337915235
+        },
+        "qwen35_2b_base": {
+          "correct": 470,
+          "accuracy": 0.5383734249713631
+        }
+      }
+    },
+    "wanli": {
+      "total": 256,
+      "models": {
+        "jev": {
+          "correct": 194,
+          "accuracy": 0.7578125
+        },
+        "kev_9b": {
+          "correct": 180,
+          "accuracy": 0.703125
+        },
+        "qev_9b": {
+          "correct": 186,
+          "accuracy": 0.7265625
+        },
+        "qwen35_9b_base": {
+          "correct": 174,
+          "accuracy": 0.6796875
+        },
+        "qev_2b": {
+          "correct": 173,
+          "accuracy": 0.67578125
+        },
+        "qwen35_2b_base": {
+          "correct": 129,
+          "accuracy": 0.50390625
+        }
+      }
+    },
+    "jevbench_public": {
+      "total": 231,
+      "models": {
+        "jev": {
+          "correct": 198,
+          "accuracy": 0.8571428571428571
+        },
+        "kev_9b": {
+          "correct": 175,
+          "accuracy": 0.7575757575757576
+        },
+        "qev_9b": {
+          "correct": 188,
+          "accuracy": 0.8138528138528138
+        },
+        "qwen35_9b_base": {
+          "correct": 175,
+          "accuracy": 0.7575757575757576
+        },
+        "qev_2b": {
+          "correct": 172,
+          "accuracy": 0.7445887445887446
+        },
+        "qwen35_2b_base": {
+          "correct": 146,
+          "accuracy": 0.6320346320346321
+        }
+      }
+    }
+  },
+  "jevbench_subsets": {
+    "original": {
+      "total": 72,
+      "correct": {
+        "jev": 71,
+        "kev_9b": 65,
+        "qev_9b": 65,
+        "qwen35_9b_base": 59,
+        "qev_2b": 64,
+        "qwen35_2b_base": 44
+      }
+    },
+    "easy": {
+      "total": 48,
+      "correct": {
+        "jev": 48,
+        "kev_9b": 48,
+        "qev_9b": 48,
+        "qwen35_9b_base": 48,
+        "qev_2b": 48,
+        "qwen35_2b_base": 48
+      }
+    },
+    "hard": {
+      "total": 111,
+      "correct": {
+        "jev": 79,
+        "kev_9b": 62,
+        "qev_9b": 75,
+        "qwen35_9b_base": 68,
+        "qev_2b": 60,
+        "qwen35_2b_base": 54
+      }
+    }
+  },
+  "notes": [
+    "Public JevBench accuracy is not the official composite score.",
+    "MMLU-Pro: Kev answered 992/1000; its 8 unanswered items count as wrong.",
+    "SemIf: the 144 handwritten questions, excluding the 108 perturbations.",
+    "Cross-model precisions and execution implementations differ.",
+    "The Qev release is one seed, and these comparisons do not isolate architecture gains.",
+    "The 2B columns use the same clean development and 144-question SemIf subsets as the 9B comparison."
+  ],
+  "research_report_sha256": "6f094e2e35217089b1ef9365525d3ddbfa46d1354e2faff4b4a25e1bbcfa63d9"
+}

evaluation.json

@@ -0,0 +1,93 @@
+{
+  "model": "Qev-2B",
+  "method": "Teacher-probability cross entropy followed by replay and representation-response distillation",
+  "candidate_preview": false,
+  "seed": 17,
+  "continuation_steps": 800,
+  "full_suites": {
+    "decision_dev": {
+      "qev_2b": {
+        "questions": 1468,
+        "correct": 1266,
+        "accuracy": 0.8623978201634878
+      },
+      "qwen35_2b_base": {
+        "questions": 1468,
+        "correct": 927,
+        "accuracy": 0.6314713896457765
+      }
+    },
+    "transfer_dev": {
+      "qev_2b": {
+        "questions": 764,
+        "correct": 584,
+        "accuracy": 0.7643979057591623
+      },
+      "qwen35_2b_base": {
+        "questions": 764,
+        "correct": 493,
+        "accuracy": 0.6452879581151832
+      }
+    },
+    "mmlupro": {
+      "qev_2b": {
+        "questions": 1000,
+        "correct": 387,
+        "accuracy": 0.387
+      },
+      "qwen35_2b_base": {
+        "questions": 1000,
+        "correct": 312,
+        "accuracy": 0.312
+      }
+    },
+    "semif": {
+      "qev_2b": {
+        "questions": 252,
+        "correct": 215,
+        "accuracy": 0.8531746031746031
+      },
+      "qwen35_2b_base": {
+        "questions": 252,
+        "correct": 152,
+        "accuracy": 0.6031746031746031
+      }
+    },
+    "scienthoon": {
+      "qev_2b": {
+        "questions": 873,
+        "correct": 628,
+        "accuracy": 0.7193585337915235
+      },
+      "qwen35_2b_base": {
+        "questions": 873,
+        "correct": 470,
+        "accuracy": 0.5383734249713631
+      }
+    },
+    "wanli": {
+      "qev_2b": {
+        "questions": 256,
+        "correct": 173,
+        "accuracy": 0.67578125
+      },
+      "qwen35_2b_base": {
+        "questions": 256,
+        "correct": 129,
+        "accuracy": 0.50390625
+      }
+    },
+    "public": {
+      "qev_2b": {
+        "questions": 231,
+        "correct": 172,
+        "accuracy": 0.7445887445887446
+      },
+      "qwen35_2b_base": {
+        "questions": 231,
+        "correct": 146,
+        "accuracy": 0.6320346320346321
+      }
+    }
+  }
+}

head.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d6adfa48de170d31cef0f38087e2da1dc4a240474fbcc849808ced719fc408d
+size 8967196

initial_training_config.json

@@ -0,0 +1,55 @@
+{
+  "model": {
+    "base": "Qwen/Qwen3.5-2B-Base",
+    "revision": "b1485b2fa6dfa1287294f269f5fb618e03d52d7c",
+    "head_dim": 256,
+    "head_heads": 4,
+    "head_layers": 2,
+    "lora_rank": 64,
+    "lora_dropout": 0.0,
+    "weights_dtype": "bf16",
+    "attention": "sdpa",
+    "rows_per_forward": 32,
+    "layout": "state-question-candidate.v1",
+    "max_padding_ratio": 1.25,
+    "max_padded_tokens": 8192,
+    "choice_none_policy": "as-provided",
+    "candidate_interaction": "last-full-attention"
+  },
+  "limits": {
+    "max_state": 1024,
+    "max_question": 512,
+    "max_candidate": 256,
+    "max_path": 2048,
+    "max_candidates": 128
+  },
+  "training": {
+    "epochs": 2,
+    "batch_size": 8,
+    "accum": 2,
+    "seed": 17,
+    "lr": 2e-05,
+    "head_lr": 0.0001,
+    "weight_decay": 0.01,
+    "warmup_steps": 20,
+    "head_warmup_steps": 100,
+    "gradient_checkpointing": true,
+    "save_every": 100,
+    "autocast": "bf16",
+    "batch_assignment": "leaf-balanced-v1",
+    "prefix_execution": "tree-batched",
+    "joint_gate_lr": 0.01,
+    "none_insert_prob": 0.2,
+    "none_insert_absent_frac": 0.5,
+    "none_insert_exempt_sources": [
+      "jev_distill/"
+    ],
+    "late_split": "late_train",
+    "late_fraction": 0.5,
+    "late_repeats": 3,
+    "distillation": {
+      "cache": "data/teacher-logits",
+      "weight": 1.0
+    }
+  }
+}

joint.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6801fc3c2569c8a99e7d5471f7a25a34b558dbd1dcf2b7c5dea1de241cb2046d
+size 104

licenses/Crafter-MIT.txt

@@ -0,0 +1,17 @@
+Copyright (c) 2021 Danijar Hafner
+
+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.

licenses/JevBench-MIT.txt

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Florian Standhartinger and contributors
+
+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.

licenses/JevBench-THIRD-PARTY.md

@@ -0,0 +1,51 @@
+# Third-party projects, data and services
+
+MIT (see `LICENSE`) covers this harness and the 72 original public decisions in
+`datasets/public/original.jsonl`. Nothing else in this list is ours to license.
+
+## Systems evaluated
+
+Each was reached through the interface its author published. We used public
+endpoints as ordinary clients, at one request at a time, and never sent a
+provider's API key to anyone else's endpoint.
+
+| System | Author | Source |
+|---|---|---|
+| Jev 1.13.0 | TypeSafe AI | <https://docs.typesafe.ai> |
+| openjev-sglang | ekzhang | <https://github.com/ekzhang/openjev-sglang> |
+| system-one-open | mithalouni | <https://github.com/mithalouni/system-one-open> |
+| open-alternative-jev | IkerMoel | <https://github.com/ikermoel/open-alternative-jev>, <https://huggingface.co/spaces/IkerMoel/open-alternative-jev> |
+| open-jev-deberta-v3-large | Kotoba Labs | <https://huggingface.co/com-kotobalabs/open-jev-deberta-v3-large>, <https://github.com/kotoba-lang/typed-decisions> |
+| GPT-5.6 Luna | OpenAI | <https://platform.openai.com> |
+| Gemini 3.1 Flash-Lite | Google | <https://ai.google.dev> |
+| DeepSeek V4.1 Flash | DeepSeek | <https://api-docs.deepseek.com> |
+| Qwen3.8 27B | Qwen, served by Chutes | <https://chutes.ai> |
+
+Model weights, base models and each project's own code keep their own licences.
+A permissive licence on a repository is not a licence for the base model it
+fine-tunes, and we do not restate either.
+
+## Wire format
+
+The typed-decision request shape (`state`, `questions`, the `noul` / `choice` /
+`score` primitives) is TypeSafe's public HTTP API, documented at
+<https://docs.typesafe.ai/api>. Several of the open rebuilds implement it
+deliberately, which is why one adapter reaches more than one of them. JevBench is
+not affiliated with or endorsed by TypeSafe AI.
+
+## Imported decisions
+
+146 of the 242 decisions come from our own earlier auto-router experiment
+(<https://github.com/fstandhartinger/auto-model-router>): 78 routing requests and
+68 answer-adequacy judgements whose ground truth is a deterministic grader's
+verdict on a saved answer. The upstream task text is **not** redistributed here,
+because the datasets it was drawn from keep their own terms. `datasets/manifest.json`
+pins the hashes; `scripts/import_router.py` shows exactly what was taken and what
+was excluded.
+
+## Held-out decisions
+
+24 scenarios are written by us and deliberately unpublished, so the suite cannot
+be trained on in full. Only their whole-split hash and aggregate results appear
+here. They are sent to the services being evaluated, which is not the same thing
+as being public - see the limits section of the README.

licenses/Kev-Apache-2.0.txt

@@ -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 Jared Palmer
+
+   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.

licenses/Qwen3.5-Apache-2.0.txt

@@ -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.

model.json

@@ -0,0 +1,31 @@
+{
+  "format": "qev.checkpoint.v1",
+  "layout": "state-question-candidate.v1",
+  "spec": {
+    "base": "Qwen/Qwen3.5-2B-Base",
+    "revision": "b1485b2fa6dfa1287294f269f5fb618e03d52d7c",
+    "head_dim": 256,
+    "head_heads": 4,
+    "head_layers": 2,
+    "lora_rank": 64,
+    "lora_dropout": 0.0,
+    "weights_dtype": "bf16",
+    "attention": "sdpa",
+    "rows_per_forward": 32,
+    "max_padding_ratio": 1.25,
+    "max_padded_tokens": 8192,
+    "layout": "state-question-candidate.v1",
+    "choice_none_policy": "as-provided",
+    "candidate_interaction": "last-full-attention"
+  },
+  "limits": {
+    "max_state": 1024,
+    "max_question": 512,
+    "max_candidate": 256,
+    "max_path": 2048,
+    "max_candidates": 128
+  },
+  "adapter": true,
+  "readout_precision": "fp32-head-and-joint-reductions.v1",
+  "temperature": 1.0
+}

response_training_config.json

@@ -0,0 +1,22 @@
+{
+  "steps": 800,
+  "batch_endpoints": 32,
+  "microbatch_endpoints": 16,
+  "replay_fraction": 0.75,
+  "seed": 17,
+  "lr": 1e-05,
+  "head_lr": 5e-05,
+  "gate_lr": 0.005,
+  "weight_decay": 0.01,
+  "warmup_steps": 20,
+  "minimum_lr_fraction": 0.1,
+  "gradient_clip": 1.0,
+  "save_steps": [
+    400,
+    800
+  ],
+  "teacher_temperature": 1.563437713227029,
+  "student_temperature": 1.0,
+  "response_weight": 0.1,
+  "gradient_checkpointing": true
+}

tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
+size 19989325

tokenizer/tokenizer_config.json

@@ -0,0 +1,32 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": true,
+  "local_files_only": true,
+  "model_max_length": 262144,
+  "model_specific_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|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "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+",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}