WiktorMatuszek commited on
Commit
45a013d
·
0 Parent(s):

MOSS-Image-SFT MOSS-Image-SFT-FP8-W8A8-Row-Max experimental release

Browse files
.gitattributes ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz filter=lfs diff=lfs merge=lfs -text
33
+ *.zip filter=lfs diff=lfs merge=lfs -text
34
+ *.zst filter=lfs diff=lfs merge=lfs -text
35
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ assets/AMPLIFIED_4X_DIFFERENCE_GRID.png filter=lfs diff=lfs merge=lfs -text
37
+ assets/REAL_IMAGE_COMPARISON_GRID.png filter=lfs diff=lfs merge=lfs -text
38
+ assets/STYLE_DIVERSITY_DIFF_X4_SEED_2026.png filter=lfs diff=lfs merge=lfs -text
39
+ assets/STYLE_DIVERSITY_DIFF_X4_SEED_42.png filter=lfs diff=lfs merge=lfs -text
40
+ assets/STYLE_DIVERSITY_REAL_IMAGES_SEED_2026.png filter=lfs diff=lfs merge=lfs -text
41
+ assets/STYLE_DIVERSITY_REAL_IMAGES_SEED_42.png filter=lfs diff=lfs merge=lfs -text
COMPLETE.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "checkpoint_sha256": {
3
+ "diffusion_pytorch_model-00001-of-00003.bin": "f1a8c654d94af8ab4e249def0138ac35360e7b84c0e2b8b9496fab668596f3ca",
4
+ "diffusion_pytorch_model-00002-of-00003.bin": "7cc470f6fbf0a3139e8ae626f42e92277d994d27e7b2c3fd7bf0fdfc69e962f8",
5
+ "diffusion_pytorch_model-00003-of-00003.bin": "0247530f9488b9b06c98af5c2288a35288c0f81bb1de14c83d2b03bde9f11905"
6
+ },
7
+ "selected_quantized_linears": 269,
8
+ "status": "CHECKPOINT_SHARDS_RELOADED_AND_VERIFIED",
9
+ "variant": "fp8w8a8-row-max",
10
+ "verified_all_disk_weights": true
11
+ }
LICENSE ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
QUANTIZATION_RECEIPT.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adapter": null,
3
+ "component": "transformer + original pinned text encoder/tokenizer/VAE/scheduler",
4
+ "cpu_preprocessing_only": true,
5
+ "lora": null,
6
+ "model_state_dict_keys": 463,
7
+ "nvfp4_format": null,
8
+ "original_selected_weight_gib": 9.466,
9
+ "output_model_shard_total_bytes": 5088772017,
10
+ "pure_sft": true,
11
+ "quantization": "fp8dynamic-row",
12
+ "scope": "max",
13
+ "seconds_elapsed_total": 752.47,
14
+ "selected_quantized_linears": 269,
15
+ "source": "OpenMOSS-Team/MOSS-Image-SFT",
16
+ "source_reference": "OpenMOSS-Team/MOSS-Image-SFT@fc5de0d56fc910cbca6f6625a70f5f582b26deb1",
17
+ "source_revision": "fc5de0d56fc910cbca6f6625a70f5f582b26deb1",
18
+ "status": "REAL_QUANTIZED_WEIGHTS_SERIALIZED_AND_RELOADED",
19
+ "student": null,
20
+ "torch": "2.7.0+cu128",
21
+ "torchao": "0.12.0+cu128",
22
+ "variant": "fp8w8a8-row-max",
23
+ "verified_disk_reload": true,
24
+ "weights": {
25
+ "diffusion_pytorch_model-00001-of-00003.bin": {
26
+ "bytes": 2499222115,
27
+ "sha256": "f1a8c654d94af8ab4e249def0138ac35360e7b84c0e2b8b9496fab668596f3ca"
28
+ },
29
+ "diffusion_pytorch_model-00002-of-00003.bin": {
30
+ "bytes": 2494656793,
31
+ "sha256": "7cc470f6fbf0a3139e8ae626f42e92277d994d27e7b2c3fd7bf0fdfc69e962f8"
32
+ },
33
+ "diffusion_pytorch_model-00003-of-00003.bin": {
34
+ "bytes": 94893109,
35
+ "sha256": "0247530f9488b9b06c98af5c2288a35288c0f81bb1de14c83d2b03bde9f11905"
36
+ }
37
+ },
38
+ "weights_format": "diffusers .bin shards torchao tensor subclasses (torch.load weights_only=True)"
39
+ }
README.md ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: OpenMOSS-Team/MOSS-Image-SFT
4
+ base_model_relation: quantized
5
+ pipeline_tag: text-to-image
6
+ library_name: diffusers
7
+ language:
8
+ - en
9
+ - pl
10
+ tags:
11
+ - quantization
12
+ - torchao
13
+ - diffusion-transformer
14
+ - moss-image-sft
15
+ - fp8dynamic-row
16
+ ---
17
+
18
+ # MOSS-Image-SFT — FP8-W8A8-Row-Max
19
+
20
+ ## Release status
21
+
22
+ Experimental checkpoint — **further testing required**. Packed W8A8 weights have passed serialization checks, but native original SFT50 image generation and visual quality for this variant remain unverified in the reported environment. GPU and software kernel support must be tested separately.
23
+
24
+ **Benchmark interpretation:** The [image gallery and measurements](https://huggingface.co/datasets/WiktorMatuszek/moss-zimage-quantization-benchmarks) are **reference examples for visual inspection**, not a required reproducibility package or guarantee of quality, speed, or VRAM usage on another setup.
25
+
26
+ **Original MOSS-Image-SFT DiT with genuinely packed FP8 W8A8, dynamic row weights.** The model is quantized from the original SFT checkpoint; there is **no** extra training, LoRA, distillation or architecture change.
27
+
28
+ | Property | This checkpoint |
29
+ |---|---|
30
+ | Quantization | FP8 E4M3 weights and dynamic per-row activation scaling |
31
+ | Coverage | **269 / 269** transformer Linear projections (all selected projections quantized) |
32
+ | Stored DiT weights | **4.739 GiB**, **49.9%** smaller than original 9.467 GiB BF16 DiT |
33
+ | Checkpoint files | Three TorchAO/PyTorch `.bin` weight shards; **not** generic safetensors |
34
+ | Upstream revision | [OpenMOSS-Team/MOSS-Image-SFT](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT), `fc5de0d56fc910cbca6f6625a70f5f582b26deb1` |
35
+ | Integrity | All 463 original state keys matched; shards reloaded and SHA-256 verified |
36
+
37
+ This repo contains the **transformer only**. The original Qwen text encoder, tokenizer, FP32 VAE, scheduler and pipeline code are loaded from the pinned upstream revision as needed. The stored DiT size is **not** total RAM/VRAM or whole-pipeline model size.
38
+
39
+ ## Evidence and measured limits
40
+
41
+ **Weights:** all 269 quantized Linear weight matrices compared with the original BF16 tensors: parameter-weighted MSE **1.52e-06**, SQNR **31.54 dB**, cosine **0.999649**. These are *weight-space errors*, not generated-image correctness. Detailed [per-layer measurements](benchmark/weight_quantization_metrics.json), [quantization receipt](QUANTIZATION_RECEIPT.json), [checkpoint integrity](COMPLETE.json).
42
+
43
+ **Original SFT50 image study:** **Not measured** with the original 50-step pipeline and native runtime in the recorded SM80 benchmark. Checkpoint weight verification does not imply that image quality or GPU execution passed.
44
+
45
+ **Runtime qualification:** the recorded SM80 setup had no supported native FP8 W8A8 execution for the original SFT50 test. A different runtime or short student-adapter control cannot certify this checkpoint's native performance or original-model image quality. See the [measured archive](https://huggingface.co/datasets/WiktorMatuszek/moss-zimage-quantization-benchmarks).
46
+
47
+ ## Use with original SFT inference
48
+
49
+ Install a CUDA-compatible PyTorch for your GPU, followed by the repository requirements. The conversion environment was PyTorch 2.7.0+cu128 / TorchAO 0.12.0+cu128; compatibility with packed tensor subclasses and native kernels is device dependent.
50
+
51
+ ```bash
52
+ hf download WiktorMatuszek/MOSS-Image-SFT-FP8-W8A8-Row-Max --local-dir ./moss-quant
53
+ cd moss-quant
54
+ python -m pip install -r requirements.txt
55
+
56
+ # Check all real saved shards before using the model
57
+ python generate.py --verify-only
58
+
59
+ # Original MOSS-SFT: 50 denoising steps, CFG 3, 1024px
60
+ python generate.py --prompt-file example_layout.json --output image.png
61
+ ```
62
+
63
+ The launcher downloads the **unchanged missing upstream components**, not duplicate original BF16 DiT weights. Optional [Qwen FP8 encoder](https://huggingface.co/WiktorMatuszek/MOSS-Image-SFT-Qwen3.5-FP8-WO) / [Flux2 VAE FP16](https://huggingface.co/WiktorMatuszek/MOSS-Image-SFT-VAE-FP16) can be selected with `--encoder-repo` and `--vae-repo`; that combination needs its own end-to-end quality validation.
64
+
65
+ ## Files and limitations
66
+
67
+ `transformer/` contains the quantized checkpoint; `generate.py`, `run_quantized_sft.py`, `compact_components.py` contain the strict pinned loader; `QUANTIZATION_RECEIPT.json` and `COMPLETE.json` record checksums. Frozen prompt suites are archived once in the [research dataset](https://huggingface.co/datasets/WiktorMatuszek/moss-zimage-quantization-benchmarks/tree/main/research/moss50-20261021), not copied into this model repo.
68
+
69
+ Pixel similarity does **not** prove correct Polish spelling, photorealism, aesthetic quality or native-kernel support. Test speed and GPU memory on your intended stack. The full cross-format gallery, including explicitly missing measurements, is maintained **once** in the [benchmark dataset](https://huggingface.co/datasets/WiktorMatuszek/moss-zimage-quantization-benchmarks).
70
+
71
+ **Attribution:** independently quantized from [OpenMOSS-Team/MOSS-Image-SFT](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT) at `fc5de0d56fc910cbca6f6625a70f5f582b26deb1`; original Apache-2.0 license in [LICENSE](LICENSE). Not an official OpenMOSS-Team release.
benchmark/weight_quantization_metrics.json ADDED
The diff for this file is too large to render. See raw diff
 
compact_components.py ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Strict immutable provenance gates for interchangeable MOSS-SFT components.
3
+
4
+ The combined pipeline always uses original OpenMOSS MOSS-Image-SFT architecture,
5
+ tokenizer, scheduler and sampler. Component paths are local staged weight folders
6
+ and must have a matching complete source-revision receipt; never select a model
7
+ based only on a filename, a loose remote branch, or an unrelated Qwen version.
8
+ """
9
+ from pathlib import Path
10
+ import hashlib
11
+ import json
12
+ import torch
13
+
14
+ REV="fc5de0d56fc910cbca6f6625a70f5f582b26deb1"
15
+
16
+ def sha(path):
17
+ h=hashlib.sha256()
18
+ with Path(path).open("rb") as f:
19
+ for piece in iter(lambda:f.read(4*1024*1024),b""):
20
+ h.update(piece)
21
+ return h.hexdigest()
22
+
23
+ def inspect_qwen(root,verify_sha=True):
24
+ root=Path(root).resolve()
25
+ manifest=root/"COMPLETE_ENCODER.json"
26
+ if not manifest.is_file():raise RuntimeError("Missing full encoder export receipt")
27
+ m=json.loads(manifest.read_text())
28
+ if (m.get("status")!="COMPLETE_RELOADED_REAL_TORCHAO_FP8WO_ENCODER"
29
+ or m.get("source_revision")!=REV or m.get("converted_linear_weights")!=248
30
+ or m.get("qwen_text_encoder_stored_shards")!=4
31
+ or m.get("quantization")!="Float8WeightOnlyConfig"):
32
+ raise RuntimeError("Not the validated original-Qwen3.5 FP8 weight-only encoder")
33
+ enc=root/"text_encoder"
34
+ assert_file=enc/"config.json"
35
+ if not assert_file.is_file() or sha(assert_file)!=m["config_sha256"]:
36
+ raise RuntimeError("Encoder architecture config changed")
37
+ total=0
38
+ for x in m["shards"]:
39
+ p=enc/x["file"]
40
+ if not p.is_file() or p.stat().st_size!=x["bytes"]:
41
+ raise RuntimeError("Incomplete stored Qwen FP8WO shard: "+x["file"])
42
+ if verify_sha and sha(p)!=x["sha256"]:
43
+ raise RuntimeError("Encoder stored shard SHA-256 mismatch: "+x["file"])
44
+ total+=x["bytes"]
45
+ if total!=m["total_text_encoder_bytes"]:
46
+ raise RuntimeError("Qwen FP8WO total size mismatch")
47
+ index=json.loads((enc/"pytorch_model.bin.index.json").read_text())
48
+ if len(index["weight_map"])!=m["source_keys"]:
49
+ raise RuntimeError("Qwen weight shard index incomplete")
50
+ return m
51
+
52
+ def inspect_vae(root,verify_sha=True):
53
+ root=Path(root).resolve()
54
+ p=root/"COMPLETE_VAE.json"
55
+ if not p.is_file():raise RuntimeError("Missing VAE precision export receipt")
56
+ m=json.loads(p.read_text())
57
+ if (m.get("status")!="COMPLETE_DTYPE_CAST_FROM_PINNED_ORIGINAL"
58
+ or m.get("source_revision")!=REV or m.get("dtype") not in ("fp16","bf16")):
59
+ raise RuntimeError("Not an original Flux2 VAE precision checkpoint")
60
+ vae=root/"vae"/"diffusion_pytorch_model.safetensors"
61
+ config=root/"vae"/"config.json"
62
+ if not vae.is_file() or vae.stat().st_size!=m["output_bytes"]:
63
+ raise RuntimeError("VAE weights incomplete")
64
+ if not config.is_file() or sha(config)!=m["config_sha256"]:
65
+ raise RuntimeError("VAE config changed")
66
+ if verify_sha and sha(vae)!=m["output_sha256"]:
67
+ raise RuntimeError("VAE weights SHA-256 mismatch")
68
+ return (m,torch.float16 if m["dtype"]=="fp16" else torch.bfloat16)
example_layout.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bbox_coord_space": "abs_centralized",
3
+ "bbox_format": "yxyx",
4
+ "items": [
5
+ {
6
+ "bbox": [
7
+ -195,
8
+ -360,
9
+ 195,
10
+ 360
11
+ ],
12
+ "caption": "large legible painted typography",
13
+ "label": "text",
14
+ "text": "Wróżka z Łęczycy"
15
+ }
16
+ ],
17
+ "summary": {
18
+ "canvas": "single",
19
+ "caption": "Old wooden sign on a small town street in warm golden-hour sunlight, photorealistic.",
20
+ "meta": [
21
+ "photo"
22
+ ]
23
+ }
24
+ }
generate.py ADDED
@@ -0,0 +1,131 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Standalone runner shipped with this Hugging Face model repository.
3
+
4
+ Loads real TorchAO quantized DiT shards, pins the exact original SFT encoder,
5
+ tokenizer, scheduler, model code and VAE by immutable source commit.
6
+ Never silently falls back to BF16, a 4-step model or on-the-fly quantization.
7
+ """
8
+ import argparse
9
+ import json
10
+ import os
11
+ from pathlib import Path
12
+ import sys
13
+
14
+ SOURCE_REPO = "OpenMOSS-Team/MOSS-Image-SFT"
15
+ SOURCE_SHA = "fc5de0d56fc910cbca6f6625a70f5f582b26deb1"
16
+ SRC_ALLOW_PATTERNS = [
17
+ "LICENSE", "model_index.json", "pipeline_moss_image.py",
18
+ "text_encoder/*", "tokenizer/*", "vae/*", "scheduler/*",
19
+ ]
20
+
21
+ def main():
22
+ ap=argparse.ArgumentParser(description=__doc__)
23
+ ap.add_argument("--prompt",type=str,help="Plain caption or MOSS structured layout as a JSON string")
24
+ ap.add_argument("--prompt-file",type=Path,help="File containing a UTF-8 literal prompt, optionally JSON")
25
+ ap.add_argument("--output",type=Path,default=Path("image.png"))
26
+ ap.add_argument("--source",type=Path,default=None,
27
+ help="Original MOSS-SFT components: immutable HF cache snapshot or complete local receipt")
28
+ ap.add_argument("--encoder-root",type=Path,default=None,
29
+ help="Optional local complete, SHA-verified original Qwen3.5 TorchAO FP8WO encoder")
30
+ ap.add_argument("--encoder-repo",type=str,default=None,
31
+ help="Optional Hugging Face model ID with verified FP8WO Qwen")
32
+ ap.add_argument("--vae-root",type=Path,default=None,
33
+ help="Optional original MOSS-SFT FP16/BF16 Flux2 VAE export directory")
34
+ ap.add_argument("--vae-repo",type=str,default=None,
35
+ help="Optional Hugging Face VAE precision variant repo ID")
36
+ ap.add_argument("--component-revision",type=str,default=None,
37
+ help="Optional immutable Git SHA for auxiliary HF component repos")
38
+ ap.add_argument("--gpu-index",type=str,default=None,
39
+ help="Physical nvidia-smi index, sets CUDA_VISIBLE_DEVICES before importing torch")
40
+ ap.add_argument("--seed",type=int,default=42)
41
+ ap.add_argument("--steps",type=int,default=50)
42
+ ap.add_argument("--cfg",type=float,default=3.0)
43
+ ap.add_argument("--width",type=int,default=1024)
44
+ ap.add_argument("--height",type=int,default=1024)
45
+ ap.add_argument("--experimental-high-res",action="store_true")
46
+ ap.add_argument("--allow-unsupported-device",action="store_true",
47
+ help="Try loading on unsupported hardware experimentally; kernels may still fail")
48
+ ap.add_argument("--offline",action="store_true",help="Do not access HF; use local source snapshot cache")
49
+ ap.add_argument("--verify-only",action="store_true",help="SHA-256-verify all local transformer shards and exit")
50
+ args=ap.parse_args()
51
+ if args.encoder_root is not None and args.encoder_repo is not None:
52
+ ap.error("Specify only one Qwen encoder source: --encoder-root or --encoder-repo")
53
+ if args.vae_root is not None and args.vae_repo is not None:
54
+ ap.error("Specify only one VAE source: --vae-root or --vae-repo")
55
+ if args.offline:
56
+ os.environ["HF_HUB_OFFLINE"]="1"
57
+ if args.encoder_repo:
58
+ from huggingface_hub import snapshot_download
59
+ args.encoder_root=Path(snapshot_download(
60
+ repo_id=args.encoder_repo,revision=args.component_revision,
61
+ local_files_only=args.offline))
62
+ if args.vae_repo:
63
+ from huggingface_hub import snapshot_download
64
+ args.vae_root=Path(snapshot_download(
65
+ repo_id=args.vae_repo,revision=args.component_revision,
66
+ local_files_only=args.offline))
67
+ if args.gpu_index is not None:
68
+ if not args.gpu_index.isdigit():
69
+ raise SystemExit("--gpu-index must be an integer. Set before CUDA imports.")
70
+ os.environ["CUDA_VISIBLE_DEVICES"]=args.gpu_index
71
+ from run_quantized_sft import check_artifact,main as run_original
72
+ repo=Path(__file__).resolve().parent
73
+ receipt=check_artifact(repo)
74
+ if receipt["source_revision"] != SOURCE_SHA or not receipt["pure_sft"]:
75
+ raise RuntimeError("Checkpoint is not the pinned original full MOSS SFT.")
76
+ if args.verify_only:
77
+ from compact_components import inspect_qwen, inspect_vae
78
+ verified={"status":"SHA256_VERIFIED","variant":receipt["variant"],
79
+ "layers":receipt["selected_quantized_linears"],
80
+ "encoder":"original_BF16" if args.encoder_root is None
81
+ else inspect_qwen(args.encoder_root)["quantization"],
82
+ "vae":"original_FP32" if args.vae_root is None
83
+ else inspect_vae(args.vae_root)[0]["dtype"]}
84
+ print(json.dumps(verified,ensure_ascii=False))
85
+ return
86
+ if bool(args.prompt) == bool(args.prompt_file):
87
+ ap.error("Provide exactly one of --prompt or --prompt-file")
88
+ if args.prompt_file is not None:
89
+ raw=args.prompt_file.read_text(encoding="utf-8").strip()
90
+ try:
91
+ payload=json.loads(raw)
92
+ prompt=json.dumps(payload,ensure_ascii=False,separators=(",",":")) if isinstance(payload,(dict,list)) else str(payload)
93
+ except json.JSONDecodeError:
94
+ prompt=raw
95
+ else:
96
+ prompt=args.prompt
97
+ if args.source is not None:
98
+ source=args.source.resolve()
99
+ else:
100
+ from huggingface_hub import snapshot_download
101
+ if args.offline:os.environ["HF_HUB_OFFLINE"]="1"
102
+ patterns=list(SRC_ALLOW_PATTERNS)
103
+ if args.encoder_root is not None:
104
+ patterns.remove("text_encoder/*")
105
+ patterns.append("text_encoder/config.json")
106
+ if args.vae_root is not None:
107
+ patterns.remove("vae/*")
108
+ patterns.append("vae/config.json")
109
+ print("Downloading pinned original MOSS-SFT code/tokenizer/scheduler and missing original components only; never downloading unneeded original BF16 DiT weights.",flush=True)
110
+ source=Path(snapshot_download(
111
+ repo_id=SOURCE_REPO,revision=SOURCE_SHA,
112
+ allow_patterns=patterns,
113
+ local_files_only=args.offline))
114
+ argv=["run_quantized_sft.py","--source",str(source),"--variant",str(repo),
115
+ "--prompt",prompt,"--out",str(args.output),"--seed",str(args.seed),
116
+ "--steps",str(args.steps),"--cfg",str(args.cfg),
117
+ "--width",str(args.width),"--height",str(args.height)]
118
+ if args.encoder_root is not None:
119
+ argv.extend(["--encoder-root",str(args.encoder_root.resolve())])
120
+ if args.vae_root is not None:
121
+ argv.extend(["--vae-root",str(args.vae_root.resolve())])
122
+ if args.experimental_high_res:argv.append("--experimental-high-res")
123
+ if args.allow_unsupported_device:argv.append("--allow-unsupported-device")
124
+ old=sys.argv
125
+ try:
126
+ sys.argv=argv
127
+ run_original()
128
+ finally:
129
+ sys.argv=old
130
+
131
+ if __name__=="__main__":main()
requirements.txt ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Install a CUDA PyTorch version supporting your specific GPU first.
2
+ # Quantization export environment: torch 2.7.0+cu128 / torchao 0.12.0+cu128
3
+ torchao==0.12.0
4
+ diffusers==0.38.0
5
+ transformers==5.12.1
6
+ accelerate==1.12.0
7
+ huggingface-hub>=1.31
8
+ safetensors>=0.5
9
+ Pillow>=11
10
+ packaging>=24
11
+ requests>=2.32
run_quantized_sft.py ADDED
@@ -0,0 +1,238 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python3
2
+ """Run full MOSS-Image-SFT quantized WEIGHTS (zero LoRA).
3
+
4
+ Usage: use downloaded exact original SFT components plus a quantized
5
+ transformer directory produced by create_full_sft_checkpoints.py.
6
+ MOSS teacher/default is 50 denoising steps and CFG=3; no 4-step student.
7
+
8
+ Hardware requirements depend on the quantization format, library and device kernels.
9
+ """
10
+ import argparse
11
+ import gc
12
+ import hashlib
13
+ import importlib.util
14
+ import json
15
+ import os
16
+ from pathlib import Path
17
+ import sys
18
+ import time
19
+ import torch
20
+
21
+ def sha256(path):
22
+ h=hashlib.sha256()
23
+ with open(path,"rb") as f:
24
+ for chunk in iter(lambda:f.read(4*1024*1024),b""):
25
+ h.update(chunk)
26
+ return h.hexdigest()
27
+
28
+ def pinned_loader(root):
29
+ spec=importlib.util.spec_from_file_location("moss_sft_native_model",root/"pipeline_moss_image.py")
30
+ module=importlib.util.module_from_spec(spec)
31
+ sys.modules[spec.name]=module
32
+ spec.loader.exec_module(module)
33
+ return module
34
+
35
+ def check_artifact(variant):
36
+ manifest=json.loads((variant/"QUANTIZATION_RECEIPT.json").read_text())
37
+ done=json.loads((variant/"COMPLETE.json").read_text())
38
+ verified = done.get("verified_all_disk_weights", done.get("reload_verified", False))
39
+ if not verified or done["variant"]!=manifest["variant"]:
40
+ raise RuntimeError("Incomplete or unverified quantized weight export")
41
+ if manifest["adapter"] is not None or manifest["lora"] is not None or manifest["student"] is not None:
42
+ raise RuntimeError("An adapter is present, expected full SFT only")
43
+ for fn, digest in done["checkpoint_sha256"].items():
44
+ f=variant/"transformer"/fn
45
+ if not f.is_file() or sha256(f)!=digest:
46
+ raise RuntimeError(f"Missing/modified full SFT checkpoint shard: {fn}")
47
+ return manifest
48
+
49
+
50
+ def verify_pinned_source(source, expected_revision):
51
+ """Verify original model identity for a verified local snapshot or an immutable HF cache snapshot.
52
+
53
+ A bare directory with unverified model files must not be accepted. The
54
+ official HF snapshot is identified by the exact commit in its cache path,
55
+ not a mutable branch/ref name. A local model_receipt.json is also valid.
56
+ """
57
+ source=Path(source).resolve()
58
+ receipt=source/"model_receipt.json"
59
+ if receipt.is_file():
60
+ data=json.loads(receipt.read_text())
61
+ if data.get("revision")!=expected_revision or not data.get("complete"):
62
+ raise RuntimeError("Original SFT local receipt revision/completeness mismatch")
63
+ elif not (source.name==expected_revision and source.parent.name=="snapshots"):
64
+ raise RuntimeError("Unverified original components. Use HF snapshot_download with exact revision "
65
+ +expected_revision+" or original complete local model_receipt.json")
66
+ required=("pipeline_moss_image.py","text_encoder/config.json",
67
+ "tokenizer/tokenizer_config.json","vae/config.json","scheduler/scheduler_config.json")
68
+ missing=[part for part in required if not (source/part).is_file()]
69
+ if missing:
70
+ raise RuntimeError("Original SFT source components missing: "+", ".join(missing))
71
+ return source
72
+
73
+
74
+ class StreamedQwenText:
75
+ """BF16 Qwen text encoder streamed via Accelerate; no LoRA."""
76
+ def __init__(self,root,device="cuda",cache_size=16,encoder_root=None):
77
+ from transformers import AutoModel,AutoTokenizer
78
+ from accelerate import cpu_offload
79
+ from compact_components import inspect_qwen
80
+ self.device=torch.device(device)
81
+ self.tokenizer=AutoTokenizer.from_pretrained(root/"tokenizer",local_files_only=True)
82
+ if encoder_root is None:
83
+ # Identical original unquantized tokenizer/encoder for BF16 reference.
84
+ enc=AutoModel.from_pretrained(root/"text_encoder",dtype=torch.bfloat16,
85
+ local_files_only=True).eval()
86
+ self.quantization="original_bf16"
87
+ else:
88
+ # This is real saved FP8WO tensor-subclass storage, not a BF16
89
+ # encoder dynamically quantized when loading or a cached BF16 bank.
90
+ encoder_root=Path(encoder_root).resolve()
91
+ info=inspect_qwen(encoder_root,verify_sha=True)
92
+ import torchao
93
+ import torchao.quantization
94
+ enc=AutoModel.from_pretrained(encoder_root/"text_encoder",
95
+ dtype=torch.bfloat16,use_safetensors=False,local_files_only=True).eval()
96
+ real=sum(type(m.weight).__name__=="AffineQuantizedTensor"
97
+ for n,m in enc.named_modules()
98
+ if n.startswith("language_model.") and isinstance(m,torch.nn.Linear))
99
+ if real!=info["converted_linear_weights"]:
100
+ raise RuntimeError(f"Packed Qwen was not actually rehydrated: {real}")
101
+ self.quantization="original_qwen3.5_fp8_weight_only_248_linear"
102
+ if hasattr(enc,"visual"):
103
+ # Image input is not used by the MOSS-SFT text-only caption path.
104
+ del enc.visual
105
+ gc.collect()
106
+ # Hold original CPU storage directly, not 15+ GiB of cloned duplicate
107
+ # Qwen weights. Accelerate replaces module slots with meta after the
108
+ # state_dict tensors have been captured by its CPU weights_map.
109
+ weights={k:v.detach() for k,v in enc.state_dict().items()}
110
+ cpu_offload(enc,execution_device=self.device,offload_buffers=True,
111
+ state_dict=weights)
112
+ del weights
113
+ self.enc=enc
114
+ self.cache={}
115
+ self.cache_size=cache_size
116
+
117
+ @torch.no_grad()
118
+ def encode(self,prompt):
119
+ if prompt in self.cache:
120
+ return self.cache[prompt]
121
+ formatted=self.tokenizer.apply_chat_template(
122
+ [{"role":"user","content":prompt}],tokenize=False,
123
+ add_generation_prompt=True,enable_thinking=True)
124
+ inp=self.tokenizer(formatted,return_tensors="pt",truncation=True,
125
+ max_length=4096).to(self.device)
126
+ result=self.enc(**inp,use_cache=False,return_dict=True)
127
+ embedding=result.last_hidden_state[:1]
128
+ data=(embedding,torch.ones(embedding.shape[:2],
129
+ device=embedding.device,dtype=torch.bool))
130
+ if len(self.cache)>=self.cache_size:
131
+ self.cache.clear()
132
+ self.cache[prompt]=data
133
+ return data
134
+
135
+
136
+ def main():
137
+ ap=argparse.ArgumentParser()
138
+ ap.add_argument("--variant",type=Path,required=True,help="folder with transformer/*.bin and COMPLETE.json")
139
+ ap.add_argument("--source",type=Path,required=True,
140
+ help="original pinned MOSS-Image-SFT, for tokenizer/text encoder/VAE/scheduler")
141
+ ap.add_argument("--prompt",required=True)
142
+ ap.add_argument("--out",type=Path,required=True)
143
+ ap.add_argument("--seed",type=int,default=123)
144
+ ap.add_argument("--steps",type=int,default=50)
145
+ ap.add_argument("--cfg",type=float,default=3.0)
146
+ ap.add_argument("--width",type=int,default=1024)
147
+ ap.add_argument("--height",type=int,default=1024)
148
+ ap.add_argument("--encoder-root",type=Path,default=None,
149
+ help="Optional separately packaged original Qwen3.5 FP8WO encoder; validates all four shards")
150
+ ap.add_argument("--vae-root",type=Path,default=None,
151
+ help="Optional locally verified original Flux2 VAE FP16 or BF16 component")
152
+ ap.add_argument("--experimental-high-res",action="store_true")
153
+ ap.add_argument("--allow-unsupported-device",action="store_true",
154
+ help="Experimental override for unsupported hardware; does NOT guarantee execution")
155
+ args=ap.parse_args()
156
+ if args.steps<1 or not 0<=args.cfg<=10:
157
+ raise ValueError("invalid inference steps/guidance")
158
+ if args.width%16 or args.height%16 or min(args.width,args.height)<512:
159
+ raise ValueError("width and height must be multiples of 16, at least 512")
160
+ if (args.height,args.width)!=(1024,1024) and not args.experimental_high_res:
161
+ raise ValueError("non-1024 sizes require --experimental-high-res (not quality-certified)")
162
+ if not torch.cuda.is_available():
163
+ raise RuntimeError("This inference CLI needs a CUDA GPU")
164
+ gpu=torch.cuda.get_device_name(0)
165
+ cc=torch.cuda.get_device_capability(0)
166
+
167
+ args.source=args.source.resolve()
168
+ args.variant=args.variant.resolve()
169
+ receipt=check_artifact(args.variant)
170
+ method=receipt.get("quantization")
171
+ required_cc=((8,9) if method in ("fp8dynamic-row","fp8dynamic-tensor")
172
+ else (10,0) if method=="nvfp4" else (0,0))
173
+ if cc < required_cc and not args.allow_unsupported_device:
174
+ raise RuntimeError(f"GPU {gpu} compute capability {cc} below minimum for {method} ({required_cc}); "
175
+ "use a compatible accelerator/runtime or explicit experimental override.")
176
+ verify_pinned_source(args.source, receipt["source_revision"])
177
+ import torchao # noqa: F401; register tensor subclasses for torch.load
178
+ import torchao.quantization # noqa: F401; activate tensor-subclass safe globals
179
+ import torchao.prototype.mx_formats # noqa: F401; NVFP4Tensor registration
180
+ from diffusers import AutoencoderKLFlux2
181
+ module=pinned_loader(args.source)
182
+ start=time.perf_counter()
183
+ transformer=module.MOSSImageTransformer.from_pretrained(
184
+ args.variant/"transformer",torch_dtype=torch.bfloat16,
185
+ use_safetensors=False,local_files_only=True).eval()
186
+ selected=receipt.get("selected_quantized_linears",
187
+ receipt.get("conversion",{}).get("selected_linear_count"))
188
+ if selected is None or selected < 200:
189
+ raise RuntimeError("Quantized weight count missing or implausibly small")
190
+ converted=sum(type(m.weight).__name__!="Parameter" for m in transformer.modules()
191
+ if isinstance(m,torch.nn.Linear))
192
+ if converted!=selected:
193
+ raise RuntimeError(f"Rehydration mismatch: {converted} != {selected}; will not fallback")
194
+ transformer=transformer.to("cuda")
195
+ print(json.dumps({"loaded_gpu":gpu, "sm":cc, "variant":receipt["variant"],
196
+ "selected":converted, "load_seconds":round(time.perf_counter()-start,2)}),flush=True)
197
+ # Qwen3.5 text encoder stays in host RAM, streams to GPU module by module.
198
+ # This is slow per new prompt, but prevents a resident >13 GiB encoder.
199
+ text=StreamedQwenText(args.source,encoder_root=args.encoder_root)
200
+ positive=text.encode(args.prompt)
201
+ negative=text.encode("") if args.cfg!=1.0 else None
202
+ if args.vae_root is None:
203
+ vae=AutoencoderKLFlux2.from_pretrained(args.source/"vae",
204
+ torch_dtype=torch.float32,local_files_only=True).eval().to("cuda")
205
+ vae_kind="original_fp32"
206
+ else:
207
+ from compact_components import inspect_vae
208
+ vae_info,dtype=inspect_vae(args.vae_root,verify_sha=True)
209
+ vae=AutoencoderKLFlux2.from_pretrained(Path(args.vae_root)/"vae",
210
+ torch_dtype=dtype,local_files_only=True).eval().to("cuda")
211
+ vae_kind=vae_info["dtype"]
212
+ scheduler=module.MOSSImageScheduler.from_pretrained(args.source/"scheduler",
213
+ local_files_only=True)
214
+ pipe=module.MOSSImagePipeline(transformer,scheduler,None,None,vae)
215
+ generator=torch.Generator("cpu").manual_seed(args.seed)
216
+ kw={}
217
+ if negative is not None:
218
+ kw["negative_prompt_embeds"]=negative[0]
219
+ kw["negative_prompt_attention_mask"]=negative[1]
220
+ torch.cuda.reset_peak_memory_stats()
221
+ torch.cuda.synchronize()
222
+ t0=time.perf_counter()
223
+ with torch.inference_mode():
224
+ im=pipe(prompt_embeds=positive[0],prompt_attention_mask=positive[1],
225
+ width=args.width,height=args.height,
226
+ num_inference_steps=args.steps,guidance_scale=args.cfg,
227
+ uncond_skip_layers=[22],generator=generator,**kw).images[0]
228
+ torch.cuda.synchronize()
229
+ args.out.parent.mkdir(parents=True,exist_ok=True)
230
+ im.save(args.out)
231
+ print(json.dumps({"image":str(args.out),"variant":receipt["variant"],"lora":False,
232
+ "encoder_precision":text.quantization,"vae_precision":vae_kind,
233
+ "seconds_image":round(time.perf_counter()-t0,2),
234
+ "peak_cuda_alloc_gib":round(torch.cuda.max_memory_allocated()/2**30,3),
235
+ "peak_cuda_reserved_gib":round(torch.cuda.max_memory_reserved()/2**30,3)}),flush=True)
236
+
237
+ if __name__=="__main__":
238
+ main()
transformer/config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "MOSSImageTransformer",
3
+ "_diffusers_version": "0.38.0",
4
+ "dim": 3072,
5
+ "ffn_dim": 8192,
6
+ "freq_dim": 256,
7
+ "latent_channels": 128,
8
+ "num_heads": 24,
9
+ "num_layers": 42,
10
+ "rope_base": 10000.0,
11
+ "text_dim": 4096
12
+ }
transformer/diffusion_pytorch_model-00001-of-00003.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1a8c654d94af8ab4e249def0138ac35360e7b84c0e2b8b9496fab668596f3ca
3
+ size 2499222115
transformer/diffusion_pytorch_model-00002-of-00003.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7cc470f6fbf0a3139e8ae626f42e92277d994d27e7b2c3fd7bf0fdfc69e962f8
3
+ size 2494656793
transformer/diffusion_pytorch_model-00003-of-00003.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0247530f9488b9b06c98af5c2288a35288c0f81bb1de14c83d2b03bde9f11905
3
+ size 94893109
transformer/diffusion_pytorch_model.bin.index.json ADDED
@@ -0,0 +1,470 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 5088772017
4
+ },
5
+ "weight_map": {
6
+ "final_linear_video.weight": "diffusion_pytorch_model-00003-of-00003.bin",
7
+ "final_norm_video.weight": "diffusion_pytorch_model-00003-of-00003.bin",
8
+ "layers.0.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
9
+ "layers.0.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
10
+ "layers.0.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
11
+ "layers.0.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
12
+ "layers.0.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
13
+ "layers.0.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
14
+ "layers.0.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
15
+ "layers.0.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
16
+ "layers.0.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
17
+ "layers.0.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
18
+ "layers.1.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
19
+ "layers.1.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
20
+ "layers.1.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
21
+ "layers.1.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
22
+ "layers.1.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
23
+ "layers.1.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
24
+ "layers.1.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
25
+ "layers.1.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
26
+ "layers.1.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
27
+ "layers.1.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
28
+ "layers.10.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
29
+ "layers.10.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
30
+ "layers.10.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
31
+ "layers.10.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
32
+ "layers.10.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
33
+ "layers.10.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
34
+ "layers.10.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
35
+ "layers.10.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
36
+ "layers.10.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
37
+ "layers.10.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
38
+ "layers.11.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
39
+ "layers.11.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
40
+ "layers.11.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
41
+ "layers.11.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
42
+ "layers.11.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
43
+ "layers.11.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
44
+ "layers.11.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
45
+ "layers.11.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
46
+ "layers.11.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
47
+ "layers.11.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
48
+ "layers.12.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
49
+ "layers.12.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
50
+ "layers.12.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
51
+ "layers.12.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
52
+ "layers.12.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
53
+ "layers.12.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
54
+ "layers.12.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
55
+ "layers.12.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
56
+ "layers.12.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
57
+ "layers.12.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
58
+ "layers.13.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
59
+ "layers.13.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
60
+ "layers.13.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
61
+ "layers.13.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
62
+ "layers.13.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
63
+ "layers.13.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
64
+ "layers.13.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
65
+ "layers.13.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
66
+ "layers.13.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
67
+ "layers.13.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
68
+ "layers.14.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
69
+ "layers.14.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
70
+ "layers.14.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
71
+ "layers.14.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
72
+ "layers.14.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
73
+ "layers.14.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
74
+ "layers.14.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
75
+ "layers.14.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
76
+ "layers.14.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
77
+ "layers.14.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
78
+ "layers.15.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
79
+ "layers.15.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
80
+ "layers.15.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
81
+ "layers.15.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
82
+ "layers.15.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
83
+ "layers.15.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
84
+ "layers.15.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
85
+ "layers.15.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
86
+ "layers.15.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
87
+ "layers.15.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
88
+ "layers.16.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
89
+ "layers.16.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
90
+ "layers.16.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
91
+ "layers.16.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
92
+ "layers.16.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
93
+ "layers.16.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
94
+ "layers.16.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
95
+ "layers.16.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
96
+ "layers.16.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
97
+ "layers.16.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
98
+ "layers.17.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
99
+ "layers.17.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
100
+ "layers.17.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
101
+ "layers.17.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
102
+ "layers.17.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
103
+ "layers.17.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
104
+ "layers.17.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
105
+ "layers.17.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
106
+ "layers.17.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
107
+ "layers.17.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
108
+ "layers.18.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
109
+ "layers.18.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
110
+ "layers.18.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
111
+ "layers.18.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
112
+ "layers.18.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
113
+ "layers.18.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
114
+ "layers.18.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
115
+ "layers.18.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
116
+ "layers.18.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
117
+ "layers.18.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
118
+ "layers.19.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
119
+ "layers.19.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
120
+ "layers.19.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
121
+ "layers.19.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
122
+ "layers.19.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
123
+ "layers.19.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
124
+ "layers.19.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
125
+ "layers.19.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
126
+ "layers.19.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
127
+ "layers.19.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
128
+ "layers.2.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
129
+ "layers.2.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
130
+ "layers.2.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
131
+ "layers.2.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
132
+ "layers.2.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
133
+ "layers.2.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
134
+ "layers.2.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
135
+ "layers.2.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
136
+ "layers.2.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
137
+ "layers.2.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
138
+ "layers.20.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
139
+ "layers.20.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
140
+ "layers.20.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
141
+ "layers.20.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
142
+ "layers.20.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
143
+ "layers.20.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
144
+ "layers.20.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
145
+ "layers.20.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
146
+ "layers.20.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
147
+ "layers.20.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
148
+ "layers.21.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
149
+ "layers.21.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
150
+ "layers.21.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
151
+ "layers.21.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
152
+ "layers.21.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
153
+ "layers.21.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
154
+ "layers.21.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
155
+ "layers.21.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
156
+ "layers.21.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
157
+ "layers.21.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
158
+ "layers.22.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
159
+ "layers.22.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
160
+ "layers.22.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
161
+ "layers.22.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
162
+ "layers.22.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
163
+ "layers.22.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
164
+ "layers.22.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
165
+ "layers.22.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
166
+ "layers.22.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
167
+ "layers.22.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
168
+ "layers.23.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
169
+ "layers.23.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
170
+ "layers.23.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
171
+ "layers.23.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
172
+ "layers.23.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
173
+ "layers.23.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
174
+ "layers.23.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
175
+ "layers.23.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
176
+ "layers.23.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
177
+ "layers.23.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
178
+ "layers.24.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
179
+ "layers.24.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
180
+ "layers.24.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
181
+ "layers.24.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
182
+ "layers.24.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
183
+ "layers.24.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
184
+ "layers.24.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
185
+ "layers.24.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
186
+ "layers.24.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
187
+ "layers.24.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
188
+ "layers.25.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
189
+ "layers.25.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
190
+ "layers.25.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
191
+ "layers.25.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
192
+ "layers.25.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
193
+ "layers.25.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
194
+ "layers.25.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
195
+ "layers.25.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
196
+ "layers.25.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
197
+ "layers.25.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
198
+ "layers.26.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
199
+ "layers.26.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
200
+ "layers.26.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
201
+ "layers.26.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
202
+ "layers.26.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
203
+ "layers.26.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
204
+ "layers.26.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
205
+ "layers.26.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
206
+ "layers.26.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
207
+ "layers.26.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
208
+ "layers.27.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
209
+ "layers.27.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
210
+ "layers.27.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
211
+ "layers.27.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
212
+ "layers.27.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
213
+ "layers.27.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
214
+ "layers.27.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
215
+ "layers.27.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
216
+ "layers.27.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
217
+ "layers.27.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
218
+ "layers.28.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
219
+ "layers.28.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
220
+ "layers.28.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
221
+ "layers.28.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
222
+ "layers.28.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
223
+ "layers.28.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
224
+ "layers.28.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
225
+ "layers.28.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
226
+ "layers.28.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
227
+ "layers.28.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
228
+ "layers.29.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
229
+ "layers.29.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
230
+ "layers.29.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
231
+ "layers.29.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
232
+ "layers.29.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
233
+ "layers.29.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
234
+ "layers.29.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
235
+ "layers.29.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
236
+ "layers.29.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
237
+ "layers.29.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
238
+ "layers.3.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
239
+ "layers.3.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
240
+ "layers.3.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
241
+ "layers.3.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
242
+ "layers.3.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
243
+ "layers.3.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
244
+ "layers.3.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
245
+ "layers.3.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
246
+ "layers.3.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
247
+ "layers.3.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
248
+ "layers.30.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
249
+ "layers.30.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
250
+ "layers.30.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
251
+ "layers.30.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
252
+ "layers.30.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
253
+ "layers.30.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
254
+ "layers.30.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
255
+ "layers.30.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
256
+ "layers.30.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
257
+ "layers.30.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
258
+ "layers.31.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
259
+ "layers.31.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
260
+ "layers.31.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
261
+ "layers.31.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
262
+ "layers.31.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
263
+ "layers.31.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
264
+ "layers.31.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
265
+ "layers.31.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
266
+ "layers.31.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
267
+ "layers.31.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
268
+ "layers.32.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
269
+ "layers.32.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
270
+ "layers.32.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
271
+ "layers.32.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
272
+ "layers.32.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
273
+ "layers.32.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
274
+ "layers.32.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
275
+ "layers.32.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
276
+ "layers.32.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
277
+ "layers.32.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
278
+ "layers.33.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
279
+ "layers.33.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
280
+ "layers.33.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
281
+ "layers.33.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
282
+ "layers.33.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
283
+ "layers.33.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
284
+ "layers.33.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
285
+ "layers.33.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
286
+ "layers.33.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
287
+ "layers.33.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
288
+ "layers.34.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
289
+ "layers.34.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
290
+ "layers.34.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
291
+ "layers.34.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
292
+ "layers.34.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
293
+ "layers.34.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
294
+ "layers.34.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
295
+ "layers.34.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
296
+ "layers.34.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
297
+ "layers.34.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
298
+ "layers.35.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
299
+ "layers.35.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
300
+ "layers.35.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
301
+ "layers.35.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
302
+ "layers.35.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
303
+ "layers.35.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
304
+ "layers.35.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
305
+ "layers.35.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
306
+ "layers.35.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
307
+ "layers.35.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
308
+ "layers.36.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
309
+ "layers.36.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
310
+ "layers.36.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
311
+ "layers.36.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
312
+ "layers.36.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
313
+ "layers.36.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
314
+ "layers.36.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
315
+ "layers.36.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
316
+ "layers.36.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
317
+ "layers.36.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
318
+ "layers.37.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
319
+ "layers.37.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
320
+ "layers.37.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
321
+ "layers.37.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
322
+ "layers.37.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
323
+ "layers.37.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
324
+ "layers.37.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
325
+ "layers.37.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
326
+ "layers.37.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
327
+ "layers.37.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
328
+ "layers.38.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
329
+ "layers.38.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
330
+ "layers.38.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
331
+ "layers.38.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
332
+ "layers.38.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
333
+ "layers.38.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
334
+ "layers.38.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
335
+ "layers.38.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
336
+ "layers.38.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
337
+ "layers.38.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
338
+ "layers.39.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
339
+ "layers.39.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
340
+ "layers.39.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
341
+ "layers.39.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
342
+ "layers.39.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
343
+ "layers.39.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
344
+ "layers.39.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
345
+ "layers.39.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
346
+ "layers.39.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
347
+ "layers.39.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
348
+ "layers.4.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
349
+ "layers.4.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
350
+ "layers.4.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
351
+ "layers.4.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
352
+ "layers.4.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
353
+ "layers.4.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
354
+ "layers.4.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
355
+ "layers.4.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
356
+ "layers.4.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
357
+ "layers.4.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
358
+ "layers.40.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
359
+ "layers.40.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
360
+ "layers.40.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
361
+ "layers.40.proj.ffn_down.weight": "diffusion_pytorch_model-00002-of-00003.bin",
362
+ "layers.40.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00002-of-00003.bin",
363
+ "layers.40.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
364
+ "layers.40.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
365
+ "layers.40.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
366
+ "layers.40.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
367
+ "layers.40.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
368
+ "layers.41.k_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
369
+ "layers.41.norm1.weight": "diffusion_pytorch_model-00002-of-00003.bin",
370
+ "layers.41.norm2.weight": "diffusion_pytorch_model-00002-of-00003.bin",
371
+ "layers.41.proj.ffn_down.weight": "diffusion_pytorch_model-00003-of-00003.bin",
372
+ "layers.41.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00003-of-00003.bin",
373
+ "layers.41.proj.k.weight": "diffusion_pytorch_model-00002-of-00003.bin",
374
+ "layers.41.proj.o.weight": "diffusion_pytorch_model-00002-of-00003.bin",
375
+ "layers.41.proj.q.weight": "diffusion_pytorch_model-00002-of-00003.bin",
376
+ "layers.41.proj.v.weight": "diffusion_pytorch_model-00002-of-00003.bin",
377
+ "layers.41.q_norm.weight": "diffusion_pytorch_model-00002-of-00003.bin",
378
+ "layers.5.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
379
+ "layers.5.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
380
+ "layers.5.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
381
+ "layers.5.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
382
+ "layers.5.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
383
+ "layers.5.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
384
+ "layers.5.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
385
+ "layers.5.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
386
+ "layers.5.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
387
+ "layers.5.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
388
+ "layers.6.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
389
+ "layers.6.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
390
+ "layers.6.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
391
+ "layers.6.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
392
+ "layers.6.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
393
+ "layers.6.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
394
+ "layers.6.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
395
+ "layers.6.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
396
+ "layers.6.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
397
+ "layers.6.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
398
+ "layers.7.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
399
+ "layers.7.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
400
+ "layers.7.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
401
+ "layers.7.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
402
+ "layers.7.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
403
+ "layers.7.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
404
+ "layers.7.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
405
+ "layers.7.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
406
+ "layers.7.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
407
+ "layers.7.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
408
+ "layers.8.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
409
+ "layers.8.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
410
+ "layers.8.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
411
+ "layers.8.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
412
+ "layers.8.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
413
+ "layers.8.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
414
+ "layers.8.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
415
+ "layers.8.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
416
+ "layers.8.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
417
+ "layers.8.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
418
+ "layers.9.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
419
+ "layers.9.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
420
+ "layers.9.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
421
+ "layers.9.proj.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
422
+ "layers.9.proj.ffn_up_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
423
+ "layers.9.proj.k.weight": "diffusion_pytorch_model-00001-of-00003.bin",
424
+ "layers.9.proj.o.weight": "diffusion_pytorch_model-00001-of-00003.bin",
425
+ "layers.9.proj.q.weight": "diffusion_pytorch_model-00001-of-00003.bin",
426
+ "layers.9.proj.v.weight": "diffusion_pytorch_model-00001-of-00003.bin",
427
+ "layers.9.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
428
+ "rope.bands": "diffusion_pytorch_model-00001-of-00003.bin",
429
+ "shared_block_modulation.bias": "diffusion_pytorch_model-00001-of-00003.bin",
430
+ "shared_block_modulation.weight": "diffusion_pytorch_model-00001-of-00003.bin",
431
+ "text_embedder.blocks.0.ffn_down.bias": "diffusion_pytorch_model-00001-of-00003.bin",
432
+ "text_embedder.blocks.0.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
433
+ "text_embedder.blocks.0.ffn_gate.bias": "diffusion_pytorch_model-00001-of-00003.bin",
434
+ "text_embedder.blocks.0.ffn_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
435
+ "text_embedder.blocks.0.ffn_up.bias": "diffusion_pytorch_model-00001-of-00003.bin",
436
+ "text_embedder.blocks.0.ffn_up.weight": "diffusion_pytorch_model-00001-of-00003.bin",
437
+ "text_embedder.blocks.0.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
438
+ "text_embedder.blocks.0.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
439
+ "text_embedder.blocks.0.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
440
+ "text_embedder.blocks.0.proj.bias": "diffusion_pytorch_model-00001-of-00003.bin",
441
+ "text_embedder.blocks.0.proj.weight": "diffusion_pytorch_model-00001-of-00003.bin",
442
+ "text_embedder.blocks.0.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
443
+ "text_embedder.blocks.0.qkv.bias": "diffusion_pytorch_model-00001-of-00003.bin",
444
+ "text_embedder.blocks.0.qkv.weight": "diffusion_pytorch_model-00001-of-00003.bin",
445
+ "text_embedder.blocks.1.ffn_down.bias": "diffusion_pytorch_model-00001-of-00003.bin",
446
+ "text_embedder.blocks.1.ffn_down.weight": "diffusion_pytorch_model-00001-of-00003.bin",
447
+ "text_embedder.blocks.1.ffn_gate.bias": "diffusion_pytorch_model-00001-of-00003.bin",
448
+ "text_embedder.blocks.1.ffn_gate.weight": "diffusion_pytorch_model-00001-of-00003.bin",
449
+ "text_embedder.blocks.1.ffn_up.bias": "diffusion_pytorch_model-00001-of-00003.bin",
450
+ "text_embedder.blocks.1.ffn_up.weight": "diffusion_pytorch_model-00001-of-00003.bin",
451
+ "text_embedder.blocks.1.k_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
452
+ "text_embedder.blocks.1.norm1.weight": "diffusion_pytorch_model-00001-of-00003.bin",
453
+ "text_embedder.blocks.1.norm2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
454
+ "text_embedder.blocks.1.proj.bias": "diffusion_pytorch_model-00001-of-00003.bin",
455
+ "text_embedder.blocks.1.proj.weight": "diffusion_pytorch_model-00001-of-00003.bin",
456
+ "text_embedder.blocks.1.q_norm.weight": "diffusion_pytorch_model-00001-of-00003.bin",
457
+ "text_embedder.blocks.1.qkv.bias": "diffusion_pytorch_model-00001-of-00003.bin",
458
+ "text_embedder.blocks.1.qkv.weight": "diffusion_pytorch_model-00001-of-00003.bin",
459
+ "text_embedder.proj.bias": "diffusion_pytorch_model-00001-of-00003.bin",
460
+ "text_embedder.proj.weight": "diffusion_pytorch_model-00001-of-00003.bin",
461
+ "time_embedding.0.bias": "diffusion_pytorch_model-00001-of-00003.bin",
462
+ "time_embedding.0.weight": "diffusion_pytorch_model-00001-of-00003.bin",
463
+ "time_embedding.2.bias": "diffusion_pytorch_model-00001-of-00003.bin",
464
+ "time_embedding.2.weight": "diffusion_pytorch_model-00001-of-00003.bin",
465
+ "video_embedder.bias": "diffusion_pytorch_model-00001-of-00003.bin",
466
+ "video_embedder.weight": "diffusion_pytorch_model-00001-of-00003.bin",
467
+ "video_head_modulation.bias": "diffusion_pytorch_model-00003-of-00003.bin",
468
+ "video_head_modulation.weight": "diffusion_pytorch_model-00003-of-00003.bin"
469
+ }
470
+ }