Add original Qwen2.5-VL Flow one-epoch step-1346 bundle

Exact native weights and required Teacher source; not continued-training checkpoints.

.gitattributes

@@ -33,3 +33,25 @@ 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
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00001-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00002-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00003-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00004-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00005-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00006-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00007-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00008-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00009-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00010-of-00010 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00001-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00002-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00003-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00004-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00005-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00006-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00007-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00008-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00009-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00010-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00011-of-00012 filter=lfs diff=lfs merge=lfs -text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00012-of-00012 filter=lfs diff=lfs merge=lfs -text

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/LICENSE-QWEN.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 [yyyy] [name of copyright owner]
+
+   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.

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/README.md

@@ -0,0 +1,140 @@
+# Qwen2.5-VL-7B-Instruct + Flow: original one-epoch final model
+
+This is the ORIGINAL completed single-epoch Joint checkpoint, step **1346**.
+It is NOT the later continuation at step 2692 or the intermediate step 2000.
+The weights are the exact native files used for the original 57.55% evaluation.
+
+## Repository Layout
+
+Models in `maverickrzw/MZ_model` are grouped by:
+
+```text
+<base-model>/<method>/<run-id>/<checkpoint-id>/
+```
+
+This bundle is:
+
+```text
+qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/
+```
+
+Keep later epochs, different methods, and different runs in separate folders.
+Do not replace this folder with continued-training weights.
+
+## What Is Included
+
+- `manifest.json`: checkpoint identity, original SHA256, part order and SHA256.
+- `weights/`: lossless 5 GiB byte parts of two native Lightning checkpoints.
+- `restore_checkpoints.py`: verifies and reconstructs the original checkpoints.
+- `training_recipe.json`: actual recipes and hardware; local paths are placeholders.
+- `base_model_lock.json`: the frozen upstream base-model revision and file hashes.
+- `evaluation/`: the complete original 11-benchmark summary and table.
+- `LICENSE-QWEN.txt`: Apache License 2.0, as declared by the frozen Qwen model card.
+
+The two restored files are:
+
+| File | Purpose | Original bytes |
+| --- | --- | ---: |
+| `joint-step001346.ckpt` | Final trained student | 50,664,234,029 |
+| `teacher-source-step001346.ckpt` | Source required by the native student loader | 61,475,940,532 |
+
+Total download: 112,140,174,561 bytes, approximately 104.44 GiB.
+Reassembly needs another 104.44 GiB of disk space. The parts are concatenated,
+not quantized, converted, compressed, or re-trained. Their original precision
+and checkpoint bytes are preserved.
+
+This is a **native custom-model bundle**, not a standard Transformers
+`save_pretrained` folder. Do not pass this folder directly to
+`AutoModel.from_pretrained`. The full ZeRO optimizer shards are not included;
+these files do not provide full-state optimizer resumption.
+
+## Training Identity
+
+- Code: https://github.com/wmz422/lvr_pipeline_new
+- Commit: `8e03586acfdf6ba4cdd959b25f24a78c6daca4d1`
+- Base: `Qwen/Qwen2.5-VL-7B-Instruct`
+- Base revision: `cc594898137f460bfe9f0759e9844b3ce807cfb5`
+- Training stages: Teacher 1346, Warmup 200, Joint 1346 optimizer updates.
+- Joint: one completed epoch; Lightning's saved zero-based epoch index is 0.
+- Seed: 42; deterministic latent interface: 8 x 512.
+- Actual execution: 8 GPUs, ZeRO-3, GPU optimizers, no parameter offload.
+- Teacher/Warmup/Joint microbatches: 8 / 8 / 4.
+- Accumulation: 2 / 2 / 4; regular global batch 128; original epoch tail 64.
+
+The training/model/loss/data recipe is the original one-epoch recipe.
+Hardware execution is NOT a literal copy of the default four-GPU launcher:
+eight GPUs and a CUDA Teacher optimizer were used to avoid host-memory OOM.
+The repository's model and training implementation were not changed.
+
+## Download And Restore
+
+Install the Hugging Face CLI on the downloading machine, then:
+
+```bash
+hf download maverickrzw/MZ_model \
+  --include "qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/**" \
+  --local-dir ./MZ_model
+
+BUNDLE=./MZ_model/qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346
+python "$BUNDLE/restore_checkpoints.py" --output-dir ./epoch1-checkpoints
+```
+
+For reproducibility, add `--revision <Hub commit SHA>` to `hf download`.
+No access token is embedded in this bundle. Public downloads do not need one.
+Existing different output files are never overwritten; matching files are reused.
+To verify all parts without writing reconstructed files:
+
+```bash
+python "$BUNDLE/restore_checkpoints.py" \
+  --output-dir ./unused-verification-output --verify-only
+```
+
+Only load trusted native `.ckpt` files: their format uses Python/PyTorch
+serialization. Verify this bundle's SHA256 checksums before loading.
+
+## Native Loading And Evaluation
+
+Use the pinned repository, not the latest branch:
+
+```bash
+git clone https://github.com/wmz422/lvr_pipeline_new.git lvr-one-epoch
+cd lvr-one-epoch
+git checkout 8e03586acfdf6ba4cdd959b25f24a78c6daca4d1
+```
+
+Follow that commit's README to install the pinned environment, prepare the
+Qwen base files, and (for full benchmarking) the frozen benchmark data.
+No FLUX or VAE cache is needed for predicted-latent inference.
+Use the downloaded base-model revision in `base_model_lock.json`.
+
+After restoring the checkpoints, replace the paths with your local absolute paths:
+
+```bash
+.venv/bin/python reproduce.py evaluate \
+  --mode predicted --gpus 4 \
+  --checkpoint /your/epoch1-checkpoints/joint-step001346.ckpt \
+  --source-checkpoint /your/epoch1-checkpoints/teacher-source-step001346.ckpt \
+  --expected-step 1346 \
+  --output results/qwen25-flow-epoch1
+```
+
+Both checkpoint arguments are required by the original native loader.
+The checkpoints preserve original metadata, including some original local paths;
+the explicit source/base-model overrides in the native entrypoint relocate them.
+`reproduce.py` supplies the local base-model path from its configured checkout.
+
+## Original Evaluation
+
+Complete 11-benchmark evaluation: 9,957 records, 10,107 generations.
+Macro accuracy: **57.55%** (unrounded: 57.54641442945964%).
+See `evaluation/main_table.md` and `evaluation/summary.json` for every benchmark.
+Greedy decoding, max 512 new tokens, and the original parsing/scoring policy.
+Known training/development image overlaps are retained and disclosed by the suite.
+This single run is not a statistical significance or generalization claim.
+
+## Upstream Terms
+
+The pinned Qwen2.5-VL-7B-Instruct model card declares Apache License 2.0;
+its standard license text is included as `LICENSE-QWEN.txt`.
+Consult the original code repository and the upstream Qwen model's usage terms.
+This archive does not grant additional rights over upstream assets.

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/base_model_lock.json

@@ -0,0 +1,90 @@
+{
+  "repository": "Qwen/Qwen2.5-VL-7B-Instruct",
+  "revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
+  "verified_files": [
+    {
+      "name": "chat_template.json",
+      "bytes": 1050,
+      "sha256": "ad60d90252ed0b0705ba14e2d0ad0fec0beac1ea955642b54059b36052d8bc96",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "config.json",
+      "bytes": 1374,
+      "sha256": "77d9ec7321cc572e3579e2c84799c9cadaded63c49ce93b101733349fc330c43",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "generation_config.json",
+      "bytes": 216,
+      "sha256": "0a3aea82869fe29f20dc95ccf3e2bcff380eca1f5ad6447a4a4b37110b08e43e",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "merges.txt",
+      "bytes": 1671839,
+      "sha256": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "model-00001-of-00005.safetensors",
+      "bytes": 3900233256,
+      "sha256": "e97b877e47fde53a6c6e77aafb36e58e91ee9d95c4a3eeac6f1b5c0e6a1c986e",
+      "upstream_lfs_sha256": "e97b877e47fde53a6c6e77aafb36e58e91ee9d95c4a3eeac6f1b5c0e6a1c986e"
+    },
+    {
+      "name": "model-00002-of-00005.safetensors",
+      "bytes": 3864726320,
+      "sha256": "a9a300a43b4724eee2abe7c18ceb26768d0ab011eb0cad19d9bfd2476a24d024",
+      "upstream_lfs_sha256": "a9a300a43b4724eee2abe7c18ceb26768d0ab011eb0cad19d9bfd2476a24d024"
+    },
+    {
+      "name": "model-00003-of-00005.safetensors",
+      "bytes": 3864726424,
+      "sha256": "111223d173e00bbee81cba1216fad28668df3476706b7fd26f4d5b50f8b3a507",
+      "upstream_lfs_sha256": "111223d173e00bbee81cba1216fad28668df3476706b7fd26f4d5b50f8b3a507"
+    },
+    {
+      "name": "model-00004-of-00005.safetensors",
+      "bytes": 3864733680,
+      "sha256": "ef47f634fa57d46ee134edcc09f34085a47da1e16c12a2abe0d67118be6d72ed",
+      "upstream_lfs_sha256": "ef47f634fa57d46ee134edcc09f34085a47da1e16c12a2abe0d67118be6d72ed"
+    },
+    {
+      "name": "model-00005-of-00005.safetensors",
+      "bytes": 1089994880,
+      "sha256": "0c859795ad3a627a9b95bcb762e059d5b768a4a36fdd4affeff269d93fdecc67",
+      "upstream_lfs_sha256": "0c859795ad3a627a9b95bcb762e059d5b768a4a36fdd4affeff269d93fdecc67"
+    },
+    {
+      "name": "model.safetensors.index.json",
+      "bytes": 57619,
+      "sha256": "73b333b0b16e5286ddba615d2caebcd495cf7e616f52eb217a81781393d79de9",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "preprocessor_config.json",
+      "bytes": 350,
+      "sha256": "f2058c716eef96ccaed1cc1e2d0c08306b62586d535b28d9d08e691b2fab7ca0",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "tokenizer.json",
+      "bytes": 7031645,
+      "sha256": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "tokenizer_config.json",
+      "bytes": 5702,
+      "sha256": "4abd3520120e266da84c0864fee064d1fb10806f02225911a47253dd38dc5f56",
+      "upstream_lfs_sha256": null
+    },
+    {
+      "name": "vocab.json",
+      "bytes": 2776833,
+      "sha256": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910",
+      "upstream_lfs_sha256": null
+    }
+  ]
+}

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/evaluation/main_table.md

@@ -0,0 +1,19 @@
+| Benchmark | Records | Generations | Accuracy (%) |
+| --- | ---: | ---: | ---: |
+| VStarBench | 191 | 191 | 86.39 |
+| HRBench4K | 800 | 800 | 72.38 |
+| HRBench8K | 800 | 800 | 68.12 |
+| MME-RealWorld-Lite | 1919 | 1919 | 50.76 |
+| BLINK | 1901 | 1901 | 54.51 |
+| MentisOculi | 1250 | 1250 | 51.04 |
+| VSP | 400 | 400 | 80.50 |
+| VisuLogic | 1000 | 1000 | 25.00 |
+| MIRA | 546 | 546 | 11.72 |
+| MathVista_testmini | 1000 | 1000 | 66.60 |
+| SAT | 150 | 300 | 66.00 |
+| Macro mean | | | 57.55 |
+
+Full declared splits; known training/development image overlaps are retained and disclosed in the suite audit.
+Answer parsing: answer-parsing-2026-10-26-v2; wrappers=tag_repair_v1, MathVista=numeric_v1.
+BLINK: task macro average. MathVista: official normalization with local numeric extraction when selected.
+SAT: format_compat_v1, averaged over both option orders; the author's reference score is diagnostic only.

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/evaluation/summary.json

@@ -0,0 +1,597 @@
+{
+  "status": "complete",
+  "records": 9957,
+  "generations": 10107,
+  "metrics": [
+    {
+      "benchmark": "VStarBench",
+      "records": 191,
+      "generations": 191,
+      "aggregation": "sample_mean",
+      "accuracy": 0.8638743455497382,
+      "sample_accuracy": 0.8638743455497382,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "direct_attributes": {
+          "n": 115,
+          "accuracy": 0.8521739130434782
+        },
+        "relative_position": {
+          "n": 76,
+          "accuracy": 0.881578947368421
+        }
+      }
+    },
+    {
+      "benchmark": "HRBench4K",
+      "records": 800,
+      "generations": 800,
+      "aggregation": "cycle_mean",
+      "accuracy": 0.72375,
+      "sample_accuracy": 0.72375,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "cross": {
+          "n": 400,
+          "accuracy": 0.535
+        },
+        "single": {
+          "n": 400,
+          "accuracy": 0.9125
+        }
+      },
+      "cycles": {
+        "0": {
+          "n": 200,
+          "accuracy": 0.77
+        },
+        "1": {
+          "n": 200,
+          "accuracy": 0.73
+        },
+        "2": {
+          "n": 200,
+          "accuracy": 0.705
+        },
+        "3": {
+          "n": 200,
+          "accuracy": 0.69
+        }
+      }
+    },
+    {
+      "benchmark": "HRBench8K",
+      "records": 800,
+      "generations": 800,
+      "aggregation": "cycle_mean",
+      "accuracy": 0.68125,
+      "sample_accuracy": 0.68125,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "cross": {
+          "n": 400,
+          "accuracy": 0.5275
+        },
+        "single": {
+          "n": 400,
+          "accuracy": 0.835
+        }
+      },
+      "cycles": {
+        "0": {
+          "n": 200,
+          "accuracy": 0.705
+        },
+        "1": {
+          "n": 200,
+          "accuracy": 0.68
+        },
+        "2": {
+          "n": 200,
+          "accuracy": 0.66
+        },
+        "3": {
+          "n": 200,
+          "accuracy": 0.68
+        }
+      }
+    },
+    {
+      "benchmark": "MME-RealWorld-Lite",
+      "records": 1919,
+      "generations": 1919,
+      "aggregation": "sample_mean",
+      "accuracy": 0.5075560187597707,
+      "sample_accuracy": 0.5075560187597707,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "Perception/Autonomous_Driving/attribute_motion_multipedestrians": {
+          "n": 50,
+          "accuracy": 0.34
+        },
+        "Perception/Autonomous_Driving/attribute_motion_multivehicles": {
+          "n": 50,
+          "accuracy": 0.44
+        },
+        "Perception/Autonomous_Driving/attribute_motion_pedestrain": {
+          "n": 50,
+          "accuracy": 0.36
+        },
+        "Perception/Autonomous_Driving/attribute_motion_vehicle": {
+          "n": 50,
+          "accuracy": 0.42
+        },
+        "Perception/Autonomous_Driving/attribute_visual_trafficsignal": {
+          "n": 50,
+          "accuracy": 0.48
+        },
+        "Perception/Autonomous_Driving/object_count": {
+          "n": 50,
+          "accuracy": 0.28
+        },
+        "Perception/Autonomous_Driving/objects_identify": {
+          "n": 50,
+          "accuracy": 0.72
+        },
+        "Perception/Diagram and Table/diagram": {
+          "n": 50,
+          "accuracy": 0.88
+        },
+        "Perception/Diagram and Table/table": {
+          "n": 50,
+          "accuracy": 0.88
+        },
+        "Perception/Monitoring/person/attribute/color": {
+          "n": 50,
+          "accuracy": 0.46
+        },
+        "Perception/Monitoring/person/attribute/orientation": {
+          "n": 19,
+          "accuracy": 0.2631578947368421
+        },
+        "Perception/Monitoring/person/counting": {
+          "n": 50,
+          "accuracy": 0.3
+        },
+        "Perception/Monitoring/vehicle/attribute/color": {
+          "n": 50,
+          "accuracy": 0.54
+        },
+        "Perception/Monitoring/vehicle/attribute/orientation": {
+          "n": 50,
+          "accuracy": 0.28
+        },
+        "Perception/Monitoring/vehicle/counting": {
+          "n": 50,
+          "accuracy": 0.38
+        },
+        "Perception/Monitoring/vehicle/location": {
+          "n": 50,
+          "accuracy": 0.38
+        },
+        "Perception/OCR with Complex Context/adver_and_product": {
+          "n": 50,
+          "accuracy": 0.88
+        },
+        "Perception/OCR with Complex Context/book_map_poster": {
+          "n": 50,
+          "accuracy": 0.72
+        },
+        "Perception/OCR with Complex Context/license": {
+          "n": 50,
+          "accuracy": 0.82
+        },
+        "Perception/OCR with Complex Context/phone_and_address": {
+          "n": 50,
+          "accuracy": 0.9
+        },
+        "Perception/OCR with Complex Context/text_recog": {
+          "n": 50,
+          "accuracy": 0.82
+        },
+        "Perception/Remote Sensing/color": {
+          "n": 50,
+          "accuracy": 0.6
+        },
+        "Perception/Remote Sensing/count": {
+          "n": 50,
+          "accuracy": 0.34
+        },
+        "Perception/Remote Sensing/position": {
+          "n": 50,
+          "accuracy": 0.58
+        },
+        "Reasoning/Autonomous_Driving/attention_trafficsignal": {
+          "n": 50,
+          "accuracy": 0.58
+        },
+        "Reasoning/Autonomous_Driving/prediction_intention_ego": {
+          "n": 50,
+          "accuracy": 0.3
+        },
+        "Reasoning/Autonomous_Driving/prediction_intention_pedestrian": {
+          "n": 50,
+          "accuracy": 0.36
+        },
+        "Reasoning/Autonomous_Driving/prediction_intention_vehicle": {
+          "n": 50,
+          "accuracy": 0.38
+        },
+        "Reasoning/Autonomous_Driving/relation_interaction_ego2pedestrain": {
+          "n": 50,
+          "accuracy": 0.32
+        },
+        "Reasoning/Autonomous_Driving/relation_interaction_ego2trafficsignal": {
+          "n": 50,
+          "accuracy": 0.36
+        },
+        "Reasoning/Autonomous_Driving/relation_interaction_ego2vehicle": {
+          "n": 50,
+          "accuracy": 0.18
+        },
+        "Reasoning/Autonomous_Driving/relation_interaction_other2other": {
+          "n": 50,
+          "accuracy": 0.34
+        },
+        "Reasoning/Diagram and Table/diagram": {
+          "n": 50,
+          "accuracy": 0.54
+        },
+        "Reasoning/Diagram and Table/table": {
+          "n": 50,
+          "accuracy": 0.6
+        },
+        "Reasoning/Monitoring/calculate": {
+          "n": 50,
+          "accuracy": 0.28
+        },
+        "Reasoning/Monitoring/intention": {
+          "n": 50,
+          "accuracy": 0.24
+        },
+        "Reasoning/Monitoring/property": {
+          "n": 50,
+          "accuracy": 0.62
+        },
+        "Reasoning/OCR with Complex Context/character identification": {
+          "n": 50,
+          "accuracy": 0.72
+        },
+        "Reasoning/OCR with Complex Context/scene understanding": {
+          "n": 50,
+          "accuracy": 0.76
+        }
+      }
+    },
+    {
+      "benchmark": "BLINK",
+      "records": 1901,
+      "generations": 1901,
+      "aggregation": "task_mean",
+      "accuracy": 0.5450591057149335,
+      "sample_accuracy": 0.5402419779063651,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "Art_Style": {
+          "n": 117,
+          "accuracy": 0.6324786324786325
+        },
+        "Counting": {
+          "n": 120,
+          "accuracy": 0.65
+        },
+        "Forensic_Detection": {
+          "n": 132,
+          "accuracy": 0.7878787878787878
+        },
+        "Functional_Correspondence": {
+          "n": 130,
+          "accuracy": 0.3230769230769231
+        },
+        "IQ_Test": {
+          "n": 150,
+          "accuracy": 0.22666666666666666
+        },
+        "Jigsaw": {
+          "n": 150,
+          "accuracy": 0.4866666666666667
+        },
+        "Multi-view_Reasoning": {
+          "n": 133,
+          "accuracy": 0.24060150375939848
+        },
+        "Object_Localization": {
+          "n": 122,
+          "accuracy": 0.5901639344262295
+        },
+        "Relative_Depth": {
+          "n": 124,
+          "accuracy": 0.7983870967741935
+        },
+        "Relative_Reflectance": {
+          "n": 134,
+          "accuracy": 0.3582089552238806
+        },
+        "Semantic_Correspondence": {
+          "n": 139,
+          "accuracy": 0.4028776978417266
+        },
+        "Spatial_Relation": {
+          "n": 143,
+          "accuracy": 0.8391608391608392
+        },
+        "Visual_Correspondence": {
+          "n": 172,
+          "accuracy": 0.5465116279069767
+        },
+        "Visual_Similarity": {
+          "n": 135,
+          "accuracy": 0.7481481481481481
+        }
+      }
+    },
+    {
+      "benchmark": "MentisOculi",
+      "records": 1250,
+      "generations": 1250,
+      "aggregation": "task_mean",
+      "accuracy": 0.5104,
+      "sample_accuracy": 0.5104,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "form-board": {
+          "n": 250,
+          "accuracy": 0.456
+        },
+        "hinge-folding": {
+          "n": 250,
+          "accuracy": 0.32
+        },
+        "paper-fold": {
+          "n": 250,
+          "accuracy": 0.632
+        },
+        "rushhour": {
+          "n": 250,
+          "accuracy": 0.488
+        },
+        "sliding-puzzle": {
+          "n": 250,
+          "accuracy": 0.656
+        }
+      }
+    },
+    {
+      "benchmark": "VSP",
+      "records": 400,
+      "generations": 400,
+      "aggregation": "sample_mean",
+      "accuracy": 0.805,
+      "sample_accuracy": 0.805,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "maze": {
+          "n": 400,
+          "accuracy": 0.805
+        }
+      }
+    },
+    {
+      "benchmark": "VisuLogic",
+      "records": 1000,
+      "generations": 1000,
+      "aggregation": "sample_mean",
+      "accuracy": 0.25,
+      "sample_accuracy": 0.25,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "Attribute Reasoning": {
+          "n": 82,
+          "accuracy": 0.2682926829268293
+        },
+        "Other": {
+          "n": 108,
+          "accuracy": 0.23148148148148148
+        },
+        "Positional Reasoning": {
+          "n": 136,
+          "accuracy": 0.3382352941176471
+        },
+        "Quantitative Reasoning": {
+          "n": 353,
+          "accuracy": 0.2521246458923513
+        },
+        "Spatial Reasoning": {
+          "n": 231,
+          "accuracy": 0.22077922077922077
+        },
+        "Stylistic Reasoning": {
+          "n": 90,
+          "accuracy": 0.18888888888888888
+        }
+      }
+    },
+    {
+      "benchmark": "MIRA",
+      "records": 546,
+      "generations": 546,
+      "aggregation": "sample_mean",
+      "accuracy": 0.11721611721611722,
+      "sample_accuracy": 0.11721611721611722,
+      "parse_failures": 4,
+      "token_limit_reached": 0,
+      "tasks": {
+        "billiards": {
+          "n": 21,
+          "accuracy": 0.09523809523809523
+        },
+        "convex_hull": {
+          "n": 30,
+          "accuracy": 0.1
+        },
+        "cubes_count": {
+          "n": 30,
+          "accuracy": 0.0
+        },
+        "cubes_missing": {
+          "n": 30,
+          "accuracy": 0.0
+        },
+        "defuse_a_bomb": {
+          "n": 25,
+          "accuracy": 0.2
+        },
+        "electric_charge": {
+          "n": 21,
+          "accuracy": 0.19047619047619047
+        },
+        "gear_rotation": {
+          "n": 40,
+          "accuracy": 0.525
+        },
+        "localizer": {
+          "n": 24,
+          "accuracy": 0.0
+        },
+        "mirror_clock": {
+          "n": 30,
+          "accuracy": 0.03333333333333333
+        },
+        "mirror_pattern": {
+          "n": 30,
+          "accuracy": 0.3
+        },
+        "multi_piece_puzzle": {
+          "n": 30,
+          "accuracy": 0.0
+        },
+        "overlap": {
+          "n": 30,
+          "accuracy": 0.0
+        },
+        "paper_airplane": {
+          "n": 25,
+          "accuracy": 0.12
+        },
+        "puzzle": {
+          "n": 26,
+          "accuracy": 0.3076923076923077
+        },
+        "rolling_dice_sum": {
+          "n": 26,
+          "accuracy": 0.038461538461538464
+        },
+        "rolling_dice_top": {
+          "n": 26,
+          "accuracy": 0.23076923076923078
+        },
+        "rolling_dice_two": {
+          "n": 26,
+          "accuracy": 0.0
+        },
+        "trailer_cubes_count": {
+          "n": 25,
+          "accuracy": 0.0
+        },
+        "trailer_cubes_missing": {
+          "n": 25,
+          "accuracy": 0.04
+        },
+        "unfolded_cube": {
+          "n": 26,
+          "accuracy": 0.0
+        }
+      }
+    },
+    {
+      "benchmark": "MathVista_testmini",
+      "records": 1000,
+      "generations": 1000,
+      "aggregation": "sample_mean",
+      "accuracy": 0.666,
+      "sample_accuracy": 0.666,
+      "parse_failures": 3,
+      "token_limit_reached": 0,
+      "tasks": {
+        "figure question answering": {
+          "n": 269,
+          "accuracy": 0.7286245353159851
+        },
+        "geometry problem solving": {
+          "n": 208,
+          "accuracy": 0.6346153846153846
+        },
+        "math word problem": {
+          "n": 186,
+          "accuracy": 0.7204301075268817
+        },
+        "textbook question answering": {
+          "n": 158,
+          "accuracy": 0.6455696202531646
+        },
+        "visual question answering": {
+          "n": 179,
+          "accuracy": 0.5698324022346368
+        }
+      }
+    },
+    {
+      "benchmark": "SAT",
+      "records": 150,
+      "generations": 300,
+      "aggregation": "option_order_mean",
+      "accuracy": 0.66,
+      "sample_accuracy": 0.66,
+      "parse_failures": 0,
+      "token_limit_reached": 0,
+      "tasks": {
+        "action_conseq": {
+          "n": 74,
+          "accuracy": 0.7162162162162162
+        },
+        "ego_movement": {
+          "n": 46,
+          "accuracy": 0.8695652173913043
+        },
+        "goal_aim": {
+          "n": 68,
+          "accuracy": 0.7205882352941176
+        },
+        "obj_movement": {
+          "n": 46,
+          "accuracy": 0.5652173913043478
+        },
+        "perspective": {
+          "n": 66,
+          "accuracy": 0.45454545454545453
+        }
+      },
+      "author_reference_accuracy": 0.66,
+      "reference_disagreements": 0
+    }
+  ],
+  "macro_accuracy": 0.5754641442945964,
+  "overlap_policy": "retain_full_scope_and_disclose_known_overlap",
+  "mathvista_metric": "official_normalization_local_numeric_v1",
+  "mathvista_policy": "numeric_v1",
+  "answer_parsing": {
+    "version": "answer-parsing-2026-10-26-v2",
+    "base_template_version": "answer-parsing-2026-10-26-v2",
+    "template_sha256": "26562d6a533384b11f20e9b958b392ed89de7f62fde703998d8454ece55ede14",
+    "wrapper_policy": "tag_repair_v1",
+    "mathvista_policy": "numeric_v1",
+    "answer_policy": "format_compat_v1",
+    "mira_policy": "strict"
+  },
+  "test_benchmarks_used_for_selection": false,
+  "run_config_sha256": "9ec0b38b19db8a76c2691b5a0aeaf15eee391b58e920d380eeaa834177de768a",
+  "scoring_config_sha256": "e9e7bceeb58fd9ea05c0c85104c220d15c80ae93b04a11b8c04bb2d72eee8a7d"
+}

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/manifest.json

@@ -0,0 +1,157 @@
+{
+  "format": "mz-native-checkpoint-byte-shards-v1",
+  "repository": "maverickrzw/MZ_model",
+  "path_in_repo": "qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346",
+  "base_model": "Qwen/Qwen2.5-VL-7B-Instruct",
+  "base_model_revision": "cc594898137f460bfe9f0759e9844b3ce807cfb5",
+  "source_code": {
+    "repository": "https://github.com/wmz422/lvr_pipeline_new",
+    "commit": "8e03586acfdf6ba4cdd959b25f24a78c6daca4d1"
+  },
+  "method": "flow",
+  "seed": 42,
+  "joint_epochs": 1,
+  "joint_optimizer_step": 1346,
+  "continued_training": false,
+  "part_size_limit_bytes": 5368709120,
+  "archives": [
+    {
+      "name": "joint-step001346.ckpt",
+      "bytes": 50664234029,
+      "sha256": "294f25294f4f93eb465749602f03a57a3d761ed65779006bce979eb33fa440d8",
+      "parts": [
+        {
+          "path": "weights/joint-step001346.ckpt.part-00001-of-00010",
+          "bytes": 5368709120,
+          "sha256": "5862c8a452d6f06a6d598bcd6a15e8040999713b5c954e908eb3bfce91d06efd"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00002-of-00010",
+          "bytes": 5368709120,
+          "sha256": "a59bc35dd497db70952ffb72ae6a4b0230fd4802360f25da74c3c7dbf3031c59"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00003-of-00010",
+          "bytes": 5368709120,
+          "sha256": "903f9554fc353e75d5b8bc12783edab1851068a83cb9d75d5906abed8fc00b85"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00004-of-00010",
+          "bytes": 5368709120,
+          "sha256": "526688896e833565c37da700226171d9c5bd748cfa164bab77be4cd5bf35b5c0"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00005-of-00010",
+          "bytes": 5368709120,
+          "sha256": "9d944bb176ec7c41c7a6eb6a226fe0d8137a9e2b322e7222d7b3618dfd253fcf"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00006-of-00010",
+          "bytes": 5368709120,
+          "sha256": "5cec06eb8cf97fc8d0ca881f2b9d7d7a55d37ff36c71a2e667cbf4ba17ad74ba"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00007-of-00010",
+          "bytes": 5368709120,
+          "sha256": "369b22684292a70c697df8a469ef51783eba3ab292cb379bf52c4d1ce7781c8b"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00008-of-00010",
+          "bytes": 5368709120,
+          "sha256": "828d6f3a651904a0e6fa773c4efbb055debab4fcef7bfb9d18f5b0ebcbcb083e"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00009-of-00010",
+          "bytes": 5368709120,
+          "sha256": "18b43be1702c2ff0c5fbbe59883f8072ff47b83a1640c68b1ce56a5796a1d6ba"
+        },
+        {
+          "path": "weights/joint-step001346.ckpt.part-00010-of-00010",
+          "bytes": 2345851949,
+          "sha256": "b3c76a63acc7aad798428c0bb25cf5e6f6d15257a32fb34a27960eb12f0e07cb"
+        }
+      ]
+    },
+    {
+      "name": "teacher-source-step001346.ckpt",
+      "bytes": 61475940532,
+      "sha256": "024c8ed5d544b105458549c119ee0d0a26bc209ddd4a256693163b5cbeb030ad",
+      "parts": [
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00001-of-00012",
+          "bytes": 5368709120,
+          "sha256": "a8a1329fd1f9a6e93db64c6fed0dce7be8fb4f657b5ee7105b0f069ac3109bf6"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00002-of-00012",
+          "bytes": 5368709120,
+          "sha256": "4a37fdc3f243de1fafdb6fbd8ceb6a66f03062098f664dd9c8a7728d86f04298"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00003-of-00012",
+          "bytes": 5368709120,
+          "sha256": "61dbaec38ad69da5be2eb8149f4c90ee5a2b560227ca75a91929afc23a7befee"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00004-of-00012",
+          "bytes": 5368709120,
+          "sha256": "8f0daf40f837df9e0c0267e0c37441f65011f631b485b5ed56ac710b31b7c0da"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00005-of-00012",
+          "bytes": 5368709120,
+          "sha256": "392b197c4ad38cd86f866a0ce452689499a397f03d6da2095138ff215e378731"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00006-of-00012",
+          "bytes": 5368709120,
+          "sha256": "1ce38386e3d4a00c5dc5c4f1cd6bfadfe242012e0c4ea739463d920a0b560b51"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00007-of-00012",
+          "bytes": 5368709120,
+          "sha256": "e34acae78d680974d2c80769f5060521e0798ada505a9042a2fdf9067804c63a"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00008-of-00012",
+          "bytes": 5368709120,
+          "sha256": "69c56a09bc9b0f42ea4d8b9aab5d164acdb674a42aefad95ebfdc9e484a61d11"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00009-of-00012",
+          "bytes": 5368709120,
+          "sha256": "da6041e01e393857a48077141a7534bce25f14d598fc1b2af9f7430029249470"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00010-of-00012",
+          "bytes": 5368709120,
+          "sha256": "0eab3895bdc0438d2677197d2f35520706966958e219a5fb0f358f8c9185be1e"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00011-of-00012",
+          "bytes": 5368709120,
+          "sha256": "b42fa8dce8f2cef051b0463d145750a52ea99cc4417f38f21e4c00049710dd69"
+        },
+        {
+          "path": "weights/teacher-source-step001346.ckpt.part-00012-of-00012",
+          "bytes": 2420140212,
+          "sha256": "9a90dacbb90d69a1c0b5f21b71da9472ec71573fe6260f1a14cb3ea16236a104"
+        }
+      ]
+    }
+  ],
+  "checkpoint_audit": [
+    {
+      "file": "joint-step001346.ckpt",
+      "global_step": 1346,
+      "lightning_epoch_index": 0,
+      "state_tensor_keys": 1472
+    },
+    {
+      "file": "teacher-source-step001346.ckpt",
+      "global_step": 1346,
+      "lightning_epoch_index": 0,
+      "state_tensor_keys": 1467
+    }
+  ]
+}

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/restore_checkpoints.py

@@ -0,0 +1,86 @@
+"""Restore byte-identical native checkpoints from this bundle's verified parts."""
+import argparse
+import hashlib
+import json
+import os
+from pathlib import Path
+import uuid
+
+BUFFER_BYTES = 8 * 1024 * 1024
+
+
+def sha256_file(path):
+    digest = hashlib.sha256()
+    with Path(path).open("rb") as stream:
+        for block in iter(lambda: stream.read(BUFFER_BYTES), b""):
+            digest.update(block)
+    return digest.hexdigest()
+
+
+def contained_path(root, relative):
+    path = Path(relative)
+    if path.is_absolute() or ".." in path.parts:
+        raise ValueError(f"Unsafe part path: {relative}")
+    resolved = (root / path).resolve()
+    resolved.relative_to(root.resolve())
+    return resolved
+
+
+def restore(bundle, output, verify_only=False):
+    bundle, output = Path(bundle).resolve(), Path(output).resolve()
+    manifest = json.loads((bundle / "manifest.json").read_text())
+    if manifest["format"] != "mz-native-checkpoint-byte-shards-v1":
+        raise ValueError("Unsupported bundle format")
+    if not verify_only:
+        output.mkdir(parents=True, exist_ok=True)
+    for archive in manifest["archives"]:
+        name = archive["name"]
+        if Path(name).name != name:
+            raise ValueError("Checkpoint name must be a basename")
+        target = output / name
+        if not verify_only and target.exists():
+            if target.stat().st_size != archive["bytes"] or sha256_file(target) != archive["sha256"]:
+                raise ValueError(f"Existing output differs; refusing to overwrite: {target}")
+            print(f"ALREADY_VERIFIED {name}", flush=True)
+            continue
+        temporary = target.with_name(f".{name}.assembling-{uuid.uuid4().hex}")
+        combined, total = hashlib.sha256(), 0
+        destination = None if verify_only else temporary.open("xb")
+        try:
+            for part in archive["parts"]:
+                source = contained_path(bundle, part["path"])
+                if source.stat().st_size != part["bytes"]:
+                    raise ValueError(f"Wrong part size: {source}")
+                digest = hashlib.sha256()
+                with source.open("rb") as stream:
+                    for block in iter(lambda: stream.read(BUFFER_BYTES), b""):
+                        digest.update(block)
+                        combined.update(block)
+                        total += len(block)
+                        if destination is not None:
+                            destination.write(block)
+                if digest.hexdigest() != part["sha256"]:
+                    raise ValueError(f"Part checksum mismatch: {source}")
+            if total != archive["bytes"] or combined.hexdigest() != archive["sha256"]:
+                raise ValueError(f"Reconstructed checkpoint checksum mismatch: {name}")
+            if destination is not None:
+                destination.flush()
+                os.fsync(destination.fileno())
+                destination.close()
+                destination = None
+                # A hard link publishes atomically and never replaces another user's file.
+                os.link(temporary, target)
+                temporary.unlink()
+        finally:
+            if destination is not None:
+                destination.close()
+        print(f"VERIFIED {name} {total} bytes sha256={archive['sha256']}", flush=True)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--bundle-dir", type=Path, default=Path(__file__).resolve().parent)
+    parser.add_argument("--output-dir", type=Path, required=True)
+    parser.add_argument("--verify-only", action="store_true")
+    args = parser.parse_args()
+    restore(args.bundle_dir, args.output_dir, args.verify_only)

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/training_recipe.json

@@ -0,0 +1,531 @@
+{
+  "optimizer_steps": {
+    "teacher_joint": 1346,
+    "warmup": 200,
+    "joint": 1346
+  },
+  "hardware": {
+    "global_batch": 128,
+    "epoch_tail_global_batch": 64,
+    "phases": {
+      "teacher_joint": {
+        "gpus": 8,
+        "per_gpu_batch": 8,
+        "accumulation": 2,
+        "global_batch": 128,
+        "zero_stage": 3,
+        "optimizer_device": "cuda",
+        "parameter_device": "cuda",
+        "gradient_scale_reference": 16
+      },
+      "warmup": {
+        "gpus": 8,
+        "per_gpu_batch": 8,
+        "accumulation": 2,
+        "global_batch": 128,
+        "zero_stage": 3,
+        "optimizer_device": "cuda",
+        "parameter_device": "cuda",
+        "gradient_scale_reference": 8
+      },
+      "joint": {
+        "gpus": 8,
+        "per_gpu_batch": 4,
+        "accumulation": 4,
+        "global_batch": 128,
+        "zero_stage": 3,
+        "optimizer_device": "cuda",
+        "parameter_device": "cuda",
+        "gradient_scale_reference": 8
+      }
+    },
+    "eval_gpus": 8,
+    "flow_generator_sub_batch": 2,
+    "automatic_hardware_tuning": false
+  },
+  "recipes": {
+    "teacher_joint": {
+      "seed_everything": 42,
+      "trainer": {
+        "accelerator": "gpu",
+        "devices": 8,
+        "precision": null,
+        "max_epochs": 1,
+        "max_steps": 1346,
+        "accumulate_grad_batches": 2,
+        "gradient_clip_val": 1.0,
+        "log_every_n_steps": 10,
+        "val_check_interval": 1.0,
+        "num_sanity_val_steps": 1,
+        "enable_progress_bar": false,
+        "strategy": {
+          "class_path": "lightning.pytorch.strategies.DeepSpeedStrategy",
+          "init_args": {
+            "config": "${REPO}/paper_experiments/2026-10-23-qwen25-vl-7b-plain-sft-clean/configs/deepspeed_zero3.json",
+            "exclude_frozen_parameters": false
+          }
+        },
+        "logger": [
+          {
+            "class_path": "lightning.pytorch.loggers.CSVLogger",
+            "init_args": {
+              "save_dir": "${RUN_DIR}/teacher_joint",
+              "name": "metrics",
+              "version": "csv",
+              "flush_logs_every_n_steps": 10
+            }
+          },
+          {
+            "class_path": "lightning.pytorch.loggers.TensorBoardLogger",
+            "init_args": {
+              "save_dir": "${RUN_DIR}/teacher_joint",
+              "name": "metrics",
+              "version": "tb"
+            }
+          }
+        ],
+        "callbacks": [
+          {
+            "class_path": "reconstruction_experiment.runtime.UpdateCheckpoint",
+            "init_args": {
+              "steps": [
+                673,
+                1346
+              ],
+              "filename": "candidate-{step:06d}",
+              "save_top_k": -1,
+              "save_last": "link",
+              "auto_insert_metric_name": false,
+              "dirpath": "${RUN_DIR}/teacher_joint/checkpoints"
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RecoveryCheckpoint",
+            "init_args": {
+              "filename": "recovery-{step:06d}",
+              "every_n_train_steps": 400,
+              "save_top_k": 1,
+              "save_last": false,
+              "save_on_train_epoch_end": false,
+              "auto_insert_metric_name": false,
+              "dirpath": "${RUN_DIR}/teacher_joint/checkpoints"
+            }
+          },
+          {
+            "class_path": "lightning.pytorch.callbacks.LearningRateMonitor",
+            "init_args": {
+              "logging_interval": "step"
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RestoreDataCursor",
+            "init_args": {
+              "steps_per_epoch": 1346
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RunAudit"
+          },
+          {
+            "class_path": "lvr.training.audit.HostMemoryAudit"
+          },
+          {
+            "class_path": "reconstruction_experiment.audit.NoDescriptionAudit"
+          }
+        ],
+        "limit_train_batches": 1.0,
+        "default_root_dir": "${RUN_DIR}/teacher_joint",
+        "use_distributed_sampler": false,
+        "plugins": [
+          {
+            "class_path": "reconstruction_experiment.precision.FlowDeepSpeedPrecision",
+            "init_args": {
+              "precision": "bf16-mixed",
+              "reference_accumulation": 16
+            }
+          }
+        ]
+      },
+      "model": {
+        "encoder_model_name_or_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "qa_model_name_or_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "num_latents": 8,
+        "latent_dim": 512,
+        "learning_rate": 1e-05,
+        "head_learning_rate": 0.0001,
+        "kl_weight": 0.0,
+        "kl_warmup_ratio": 0.1,
+        "lr_warmup_ratio": 0.03,
+        "gradient_checkpointing": true,
+        "seed": 42,
+        "train_qa": true,
+        "gaussian": false,
+        "qa_loss_weight": 1.0,
+        "description_weight": 0.0,
+        "encoder_uses_question": false,
+        "encoder_includes_original": true,
+        "description_image_mode": "none",
+        "weight_decay": 0.01,
+        "lr_schedule_steps": 1346,
+        "encoder_type": "qwen",
+        "reconstruction_weight": 0.0,
+        "flow_model_path": "${REPO}/doc/flow_reconstruction_parallel_20261015/weights/FLUX.2-klein-base-4B",
+        "flow_weight": 0.1,
+        "flow_mapper_hidden_size": 1024,
+        "flow_gradient_checkpointing": true,
+        "flow_protocol_sha256": "20fb369048ba99d123e1f80b10186dbf853836dca59866e3cc1bbcdd3a4cccb3",
+        "flow_generator_batch_size": 2,
+        "flow_vae_batch_size": 2,
+        "flow_latent_cache_bytes": 0
+      },
+      "data": {
+        "train_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/sft/train",
+        "val_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/sft/test",
+        "train_boxes_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/align/train",
+        "val_boxes_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/align/test",
+        "processor_name_or_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "image_root": "${REPO}/lvr_pipeline_new/data",
+        "batch_size": 8,
+        "num_workers": 4,
+        "num_latents": 8,
+        "original_pixels": 2000000,
+        "crop_pixels": 262144,
+        "max_length": 32768,
+        "seed": 42,
+        "include_encoder": true,
+        "qa_crops": false,
+        "encoder_image_mode": "question_crops",
+        "encoder_includes_original": true,
+        "include_descriptions": false,
+        "description_image_mode": "none",
+        "caption_path": null,
+        "val_num_workers": 2,
+        "prefetch_factor": 1,
+        "pin_memory": true,
+        "additional_train_path": "${REPO}/paper_experiments/2026-10-26-qwen25-no-sat-ablation/data/reasoning_train_without_sat.json",
+        "additional_val_path": "${REPO}/lvr_pipeline_new/data/processed/reasoning_v1/val",
+        "persistent_workers": true,
+        "encoder_uses_question": false,
+        "include_reconstruction": false,
+        "flow_data_root": "${REPO}/paper_experiments/2026-10-27-qwen25-flow-no-sat-4gpu/data/reconstruction_2mp",
+        "flow_protocol_sha256": "20fb369048ba99d123e1f80b10186dbf853836dca59866e3cc1bbcdd3a4cccb3",
+        "flow_latent_root": "${REPO}/paper_experiments/2026-10-27-qwen25-flow-no-sat-4gpu/data/vae_latents_fp32",
+        "flow_vae_model_path": "${REPO}/doc/flow_reconstruction_parallel_20261015/weights/FLUX.2-klein-base-4B",
+        "flow_layout_buckets": true,
+        "flow_stress_first": false
+      },
+      "exp_name": "${RUN_DIR}",
+      "version": "teacher_joint"
+    },
+    "warmup": {
+      "seed_everything": 42,
+      "trainer": {
+        "accelerator": "gpu",
+        "devices": 8,
+        "precision": null,
+        "max_epochs": 1,
+        "max_steps": 200,
+        "accumulate_grad_batches": 2,
+        "gradient_clip_val": 1.0,
+        "log_every_n_steps": 10,
+        "val_check_interval": 1.0,
+        "num_sanity_val_steps": 1,
+        "enable_progress_bar": false,
+        "strategy": {
+          "class_path": "lightning.pytorch.strategies.DeepSpeedStrategy",
+          "init_args": {
+            "config": "${REPO}/paper_experiments/2026-10-23-qwen25-vl-7b-plain-sft-clean/configs/deepspeed_zero3.json",
+            "exclude_frozen_parameters": false
+          }
+        },
+        "logger": [
+          {
+            "class_path": "lightning.pytorch.loggers.CSVLogger",
+            "init_args": {
+              "save_dir": "${RUN_DIR}/warmup",
+              "name": "metrics",
+              "version": "csv",
+              "flush_logs_every_n_steps": 10
+            }
+          },
+          {
+            "class_path": "lightning.pytorch.loggers.TensorBoardLogger",
+            "init_args": {
+              "save_dir": "${RUN_DIR}/warmup",
+              "name": "metrics",
+              "version": "tb"
+            }
+          }
+        ],
+        "callbacks": [
+          {
+            "class_path": "reconstruction_experiment.runtime.UpdateCheckpoint",
+            "init_args": {
+              "steps": [
+                100,
+                200
+              ],
+              "filename": "candidate-{step:06d}",
+              "save_top_k": -1,
+              "save_last": "link",
+              "auto_insert_metric_name": false,
+              "dirpath": "${RUN_DIR}/warmup/checkpoints"
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RecoveryCheckpoint",
+            "init_args": {
+              "filename": "recovery-{step:06d}",
+              "every_n_train_steps": 400,
+              "save_top_k": 1,
+              "save_last": false,
+              "save_on_train_epoch_end": false,
+              "auto_insert_metric_name": false,
+              "dirpath": "${RUN_DIR}/warmup/checkpoints"
+            }
+          },
+          {
+            "class_path": "lightning.pytorch.callbacks.LearningRateMonitor",
+            "init_args": {
+              "logging_interval": "step"
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RestoreDataCursor",
+            "init_args": {
+              "steps_per_epoch": 1346
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RunAudit"
+          },
+          {
+            "class_path": "lvr.training.audit.HostMemoryAudit"
+          },
+          {
+            "class_path": "reconstruction_experiment.audit.NoDescriptionAudit"
+          }
+        ],
+        "limit_train_batches": 400,
+        "default_root_dir": "${RUN_DIR}/warmup",
+        "use_distributed_sampler": false,
+        "plugins": [
+          {
+            "class_path": "reconstruction_experiment.precision.MatchedDeepSpeedPrecision",
+            "init_args": {
+              "precision": "bf16-mixed",
+              "reference_accumulation": 8
+            }
+          }
+        ]
+      },
+      "model": {
+        "source_checkpoint": "${RUN_DIR}/teacher_joint/checkpoints/student_source.ckpt",
+        "base_model_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "prediction_phase": "warmup",
+        "learning_rate": 1e-05,
+        "head_learning_rate": 0.0001,
+        "weight_decay": 0.01,
+        "lr_warmup_ratio": 0.03,
+        "lr_schedule_steps": 200,
+        "latent_weight": 1.0,
+        "cosine_weight": 1.0,
+        "qa_loss_weight": 1.0,
+        "description_weight": 0.0,
+        "energy_floor": 1e-06,
+        "gradient_checkpointing": true,
+        "seed": 42,
+        "num_loops": 1,
+        "allow_question_conditioned_teacher": false
+      },
+      "data": {
+        "train_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/sft/train",
+        "val_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/sft/test",
+        "train_boxes_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/align/train",
+        "val_boxes_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/align/test",
+        "processor_name_or_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "image_root": "${REPO}/lvr_pipeline_new/data",
+        "batch_size": 8,
+        "num_workers": 4,
+        "num_latents": 8,
+        "original_pixels": 2000000,
+        "crop_pixels": 262144,
+        "max_length": 32768,
+        "seed": 42,
+        "include_encoder": true,
+        "qa_crops": false,
+        "encoder_image_mode": "question_crops",
+        "encoder_includes_original": true,
+        "include_descriptions": false,
+        "description_image_mode": "none",
+        "caption_path": null,
+        "val_num_workers": 2,
+        "prefetch_factor": 1,
+        "pin_memory": true,
+        "additional_train_path": "${REPO}/paper_experiments/2026-10-26-qwen25-no-sat-ablation/data/reasoning_train_without_sat.json",
+        "additional_val_path": "${REPO}/lvr_pipeline_new/data/processed/reasoning_v1/val",
+        "persistent_workers": true,
+        "encoder_uses_question": false,
+        "include_reconstruction": false
+      },
+      "exp_name": "${RUN_DIR}",
+      "version": "warmup"
+    },
+    "joint": {
+      "seed_everything": 42,
+      "trainer": {
+        "accelerator": "gpu",
+        "devices": 8,
+        "precision": null,
+        "max_epochs": 1,
+        "max_steps": 1346,
+        "accumulate_grad_batches": 4,
+        "gradient_clip_val": 1.0,
+        "log_every_n_steps": 10,
+        "val_check_interval": 1.0,
+        "num_sanity_val_steps": 1,
+        "enable_progress_bar": false,
+        "strategy": {
+          "class_path": "lightning.pytorch.strategies.DeepSpeedStrategy",
+          "init_args": {
+            "config": "${REPO}/paper_experiments/2026-10-23-qwen25-vl-7b-plain-sft-clean/configs/deepspeed_zero3.json",
+            "exclude_frozen_parameters": false
+          }
+        },
+        "logger": [
+          {
+            "class_path": "lightning.pytorch.loggers.CSVLogger",
+            "init_args": {
+              "save_dir": "${RUN_DIR}/joint",
+              "name": "metrics",
+              "version": "csv",
+              "flush_logs_every_n_steps": 10
+            }
+          },
+          {
+            "class_path": "lightning.pytorch.loggers.TensorBoardLogger",
+            "init_args": {
+              "save_dir": "${RUN_DIR}/joint",
+              "name": "metrics",
+              "version": "tb"
+            }
+          }
+        ],
+        "callbacks": [
+          {
+            "class_path": "reconstruction_experiment.runtime.UpdateCheckpoint",
+            "init_args": {
+              "steps": [
+                673,
+                1346
+              ],
+              "filename": "candidate-{step:06d}",
+              "save_top_k": -1,
+              "save_last": "link",
+              "auto_insert_metric_name": false,
+              "dirpath": "${RUN_DIR}/joint/checkpoints"
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RecoveryCheckpoint",
+            "init_args": {
+              "filename": "recovery-{step:06d}",
+              "every_n_train_steps": 400,
+              "save_top_k": 1,
+              "save_last": false,
+              "save_on_train_epoch_end": false,
+              "auto_insert_metric_name": false,
+              "dirpath": "${RUN_DIR}/joint/checkpoints"
+            }
+          },
+          {
+            "class_path": "lightning.pytorch.callbacks.LearningRateMonitor",
+            "init_args": {
+              "logging_interval": "step"
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RestoreDataCursor",
+            "init_args": {
+              "steps_per_epoch": 1346
+            }
+          },
+          {
+            "class_path": "reconstruction_experiment.runtime.RunAudit"
+          },
+          {
+            "class_path": "lvr.training.audit.HostMemoryAudit"
+          },
+          {
+            "class_path": "reconstruction_experiment.audit.NoDescriptionAudit"
+          }
+        ],
+        "limit_train_batches": 1.0,
+        "default_root_dir": "${RUN_DIR}/joint",
+        "use_distributed_sampler": false,
+        "plugins": [
+          {
+            "class_path": "reconstruction_experiment.precision.MatchedDeepSpeedPrecision",
+            "init_args": {
+              "precision": "bf16-mixed",
+              "reference_accumulation": 8
+            }
+          }
+        ]
+      },
+      "model": {
+        "source_checkpoint": "${RUN_DIR}/teacher_joint/checkpoints/student_source.ckpt",
+        "base_model_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "prediction_phase": "joint",
+        "learning_rate": 1e-05,
+        "head_learning_rate": 0.0001,
+        "weight_decay": 0.01,
+        "lr_warmup_ratio": 0.03,
+        "lr_schedule_steps": 1346,
+        "latent_weight": 1.0,
+        "cosine_weight": 1.0,
+        "qa_loss_weight": 1.0,
+        "description_weight": 0.0,
+        "energy_floor": 1e-06,
+        "gradient_checkpointing": true,
+        "seed": 42,
+        "num_loops": 1,
+        "allow_question_conditioned_teacher": false
+      },
+      "data": {
+        "train_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/sft/train",
+        "val_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/sft/test",
+        "train_boxes_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/align/train",
+        "val_boxes_path": "${REPO}/paper_experiments/datasets/vstar_decontaminated_v1/align/test",
+        "processor_name_or_path": "${REPO}/paper_experiments/models/Qwen2.5-VL-7B-Instruct",
+        "image_root": "${REPO}/lvr_pipeline_new/data",
+        "batch_size": 4,
+        "num_workers": 4,
+        "num_latents": 8,
+        "original_pixels": 2000000,
+        "crop_pixels": 262144,
+        "max_length": 32768,
+        "seed": 42,
+        "include_encoder": true,
+        "qa_crops": false,
+        "encoder_image_mode": "question_crops",
+        "encoder_includes_original": true,
+        "include_descriptions": false,
+        "description_image_mode": "none",
+        "caption_path": null,
+        "val_num_workers": 2,
+        "prefetch_factor": 1,
+        "pin_memory": true,
+        "additional_train_path": "${REPO}/paper_experiments/2026-10-26-qwen25-no-sat-ablation/data/reasoning_train_without_sat.json",
+        "additional_val_path": "${REPO}/lvr_pipeline_new/data/processed/reasoning_v1/val",
+        "persistent_workers": true,
+        "encoder_uses_question": false,
+        "include_reconstruction": false
+      },
+      "exp_name": "${RUN_DIR}",
+      "version": "joint",
+      "init_checkpoint": "${RUN_DIR}/warmup/checkpoints/last_full.ckpt"
+    }
+  },
+  "note": "Native default one-epoch recipe; 8-GPU CUDA-optimizer execution, not default four-GPU hardware."
+}

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00001-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5862c8a452d6f06a6d598bcd6a15e8040999713b5c954e908eb3bfce91d06efd
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00002-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a59bc35dd497db70952ffb72ae6a4b0230fd4802360f25da74c3c7dbf3031c59
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00003-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:903f9554fc353e75d5b8bc12783edab1851068a83cb9d75d5906abed8fc00b85
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00004-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:526688896e833565c37da700226171d9c5bd748cfa164bab77be4cd5bf35b5c0
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00005-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d944bb176ec7c41c7a6eb6a226fe0d8137a9e2b322e7222d7b3618dfd253fcf
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00006-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5cec06eb8cf97fc8d0ca881f2b9d7d7a55d37ff36c71a2e667cbf4ba17ad74ba
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00007-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:369b22684292a70c697df8a469ef51783eba3ab292cb379bf52c4d1ce7781c8b
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00008-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:828d6f3a651904a0e6fa773c4efbb055debab4fcef7bfb9d18f5b0ebcbcb083e
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00009-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18b43be1702c2ff0c5fbbe59883f8072ff47b83a1640c68b1ce56a5796a1d6ba
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/joint-step001346.ckpt.part-00010-of-00010

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3c76a63acc7aad798428c0bb25cf5e6f6d15257a32fb34a27960eb12f0e07cb
+size 2345851949

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00001-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8a1329fd1f9a6e93db64c6fed0dce7be8fb4f657b5ee7105b0f069ac3109bf6
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00002-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a37fdc3f243de1fafdb6fbd8ceb6a66f03062098f664dd9c8a7728d86f04298
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00003-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61dbaec38ad69da5be2eb8149f4c90ee5a2b560227ca75a91929afc23a7befee
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00004-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f0daf40f837df9e0c0267e0c37441f65011f631b485b5ed56ac710b31b7c0da
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00005-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:392b197c4ad38cd86f866a0ce452689499a397f03d6da2095138ff215e378731
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00006-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ce38386e3d4a00c5dc5c4f1cd6bfadfe242012e0c4ea739463d920a0b560b51
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00007-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e34acae78d680974d2c80769f5060521e0798ada505a9042a2fdf9067804c63a
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00008-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69c56a09bc9b0f42ea4d8b9aab5d164acdb674a42aefad95ebfdc9e484a61d11
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00009-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da6041e01e393857a48077141a7534bce25f14d598fc1b2af9f7430029249470
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00010-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0eab3895bdc0438d2677197d2f35520706966958e219a5fb0f358f8c9185be1e
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00011-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b42fa8dce8f2cef051b0463d145750a52ea99cc4417f38f21e4c00049710dd69
+size 5368709120

qwen2.5-vl-7b-instruct/flow/run-20261031-seed42/joint-epoch1-step001346/weights/teacher-source-step001346.ckpt.part-00012-of-00012

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a90dacbb90d69a1c0b5f21b71da9472ec71573fe6260f1a14cb3ea16236a104
+size 2420140212