Upload folder using huggingface_hub

.gitattributes

@@ -35,3 +35,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 sym_smoothquant/qwen_fp4_exclude_none_algo_smoothquant.png filter=lfs diff=lfs merge=lfs -text
 sym_max/qwen_fp4_exclude_none_algo_max.png filter=lfs diff=lfs merge=lfs -text
+asym_max/qwen_fp4_exclude_none_algo_max_asymmetric.png filter=lfs diff=lfs merge=lfs -text

asym_max/inference.log

@@ -0,0 +1,61 @@
+/workspace/Model-Optimizer/.venv/lib/python3.12/site-packages/torch/cuda/__init__.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
+  import pynvml  # type: ignore[import]
+Multiple distributions found for package modelopt. Picked distribution: nvidia-modelopt
+
+================================================================================
+  NVFP4_GEMM KERNEL STATUS
+================================================================================
+  ❌ NOT AVAILABLE
+     Reason: TensorRT-LLM not installed: libmpi.so.40: cannot open shared object file: No such file or directory
+
+     If you use --restore-from with a COMPRESSED checkpoint,
+     this script will EXIT with an error.
+
+     Use FAKE quantized checkpoints (without --compress) instead
+     for quality testing at BF16 speed.
+================================================================================
+
+[22:35:44] GPU: NVIDIA B200 (178.4 GB)
+[22:35:44] Loading Qwen-Image-2512 pipeline from HuggingFace...
+
+
+
+
+
+
+[22:35:45] Pipeline components:
+[22:35:45]   - vae: AutoencoderKLQwenImage (126,892,531 params)
+[22:35:45]   - text_encoder: Qwen2_5_VLForConditionalGeneration (8,292,166,656 params)
+[22:35:45]   - tokenizer: Qwen2Tokenizer
+[22:35:45]   - transformer: QwenImageTransformer2DModel (20,430,401,088 params)
+[22:35:45]   - scheduler: FlowMatchEulerDiscreteScheduler
+[22:35:45] Backbone model: QwenImageTransformer2DModel
+[22:35:45] Restoring quantized checkpoint from ./experiments_v3_asymmetric/exclude_none_algo_max_asymmetric/qwen_fp4_exclude_none_algo_max_asymmetric.pt...
+Inserted 2838 quantizers
+[22:36:03] Checkpoint restored in 18.6s
+[22:36:03] Moving pipeline to CUDA...
+[22:36:09] Pipeline ready!
+[22:36:09] 
+[22:36:09] ============================================================
+[22:36:09] ℹ️  Running with FAKE QUANTIZATION
+[22:36:09]    Quality = same as real FP4
+[22:36:09]    Speed = BF16 (no speedup)
+[22:36:09]    This is fine for quality evaluation.
+[22:36:09] ============================================================
+[22:36:09] 
+Generating image with prompt: 'a photo of an astronaut riding a horse on mars, highly detailed, 4k'
+[22:36:09] Generating image with prompt: 'a photo of an astronaut riding a horse on mars, highly detai...'
+[22:36:09] Using negative prompt: '低分辨率，低画质，肢体畸形，手指畸形，画面过饱和，蜡像感，人脸无细节，过度光滑，...'
+guidance_scale is passed as 4.0, but ignored since the model is not guidance-distilled.
+
  0%|          | 0/30 [00:00<?, ?it/s]Loading extension modelopt_cuda_ext_fp8...
+Loaded extension modelopt_cuda_ext_fp8 in 0.0 seconds
+  Step 1/30
+
  3%|▎         | 1/30 [00:01<00:32,  1.12s/it]
  7%|▋         | 2/30 [00:01<00:20,  1.38it/s]
 10%|█         | 3/30 [00:02<00:16,  1.67it/s]
 13%|█▎        | 4/30 [00:02<00:13,  1.86it/s]
 17%|█▋        | 5/30 [00:02<00:12,  1.98it/s]  Step 6/30
+
 20%|██        | 6/30 [00:03<00:11,  2.06it/s]
 23%|██▎       | 7/30 [00:03<00:10,  2.11it/s]
 27%|██▋       | 8/30 [00:04<00:10,  2.15it/s]
 30%|███       | 9/30 [00:04<00:09,  2.18it/s]
 33%|███▎      | 10/30 [00:05<00:09,  2.19it/s]  Step 11/30
+
 37%|███▋      | 11/30 [00:05<00:08,  2.21it/s]
 40%|████      | 12/30 [00:06<00:08,  2.21it/s]
 43%|████▎     | 13/30 [00:06<00:07,  2.22it/s]
 47%|████▋     | 14/30 [00:06<00:07,  2.22it/s]
 50%|█████     | 15/30 [00:07<00:06,  2.23it/s]  Step 16/30
+
 53%|█████▎    | 16/30 [00:07<00:06,  2.23it/s]
 57%|█████▋    | 17/30 [00:08<00:05,  2.23it/s]
 60%|██████    | 18/30 [00:08<00:05,  2.23it/s]
 63%|██████▎   | 19/30 [00:09<00:04,  2.23it/s]
 67%|██████▋   | 20/30 [00:09<00:04,  2.23it/s]  Step 21/30
+
 70%|███████   | 21/30 [00:10<00:04,  2.23it/s]
 73%|███████▎  | 22/30 [00:10<00:03,  2.23it/s]
 77%|███████▋  | 23/30 [00:10<00:03,  2.23it/s]
 80%|████████  | 24/30 [00:11<00:02,  2.23it/s]
 83%|████████▎ | 25/30 [00:11<00:02,  2.23it/s]  Step 26/30
+
 87%|████████▋ | 26/30 [00:12<00:01,  2.23it/s]
 90%|█████████ | 27/30 [00:12<00:01,  2.23it/s]
 93%|█████████▎| 28/30 [00:13<00:00,  2.24it/s]
 97%|█████████▋| 29/30 [00:13<00:00,  2.24it/s]  Step 30/30
+
+[22:36:23] Image generated in 14.8s, saved as ./experiments_v3_asymmetric/exclude_none_algo_max_asymmetric/qwen_fp4_exclude_none_algo_max_asymmetric.png
+[22:36:23] Done!

asym_max/qwen_fp4_exclude_none_algo_max_asymmetric.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:215300c35e5a0f2b7097896bce37f7802854f4b3ba25ef38e53a50e49d9303a2
+size 40875159450