| dev4 sweep tau=0.012 frames=120 -> /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749 |
| ========== [dev4] tau0.012_max_w3_da0.5_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 08:40:25.888062994 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 08:40:25,341 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 08:40:25,341 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 08:40:25,341 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.5, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'max'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = max |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.05s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.17s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.15s/it] |
| [2026-06-14 08:40:46,090 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 08:40:46,090 - INFO] Build DiTModel successfully |
| [2026-06-14 08:40:46,091 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 08:40:46,091 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 157.34it/s] |
| [2026-06-14 08:40:48,384 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 08:40:48,385 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 08:40:48,502 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:40:48,505 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:40:48,617 - INFO] Load checkpoint successfully |
| [2026-06-14 08:40:48,617 - INFO] Begin to generate per chunk |
| [2026-06-14 08:40:48,617 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 08:40:48,661 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:36<06:24, 96.21s/it]
InferBatch 0: 40%|████ | 2/5 [03:11<04:47, 95.92s/it]
InferBatch 0: 60%|██████ | 3/5 [04:08<02:35, 77.93s/it]
InferBatch 0: 80%|████████ | 4/5 [04:52<01:04, 64.64s/it]
InferBatch 0: 100%|██████████| 5/5 [05:17<00:00, 50.18s/it]
InferBatch 0: 100%|██████████| 5/5 [05:19<00:00, 63.94s/it] |
| [2026-06-14 08:46:10,518 - INFO] Finish MagiPipeline, max memory allocated: 23.36 GB, max memory reserved: 24.72 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_max_w3_da0.5_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.36 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.07 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.2151dB reuse=13.75% time=356s |
| ========== [dev4] tau0.012_max_w5_da0.5_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 08:46:26.478372248 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 08:46:26,932 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 08:46:26,932 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 08:46:26,932 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.5, 'detail_window_size': 5, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'max'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 5 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = max |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.33s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:19<00:00, 9.69s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:19<00:00, 9.64s/it] |
| [2026-06-14 08:46:48,427 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 08:46:48,427 - INFO] Build DiTModel successfully |
| [2026-06-14 08:46:48,428 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 08:46:48,428 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 162.07it/s] |
| [2026-06-14 08:46:50,771 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 08:46:50,771 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 08:46:50,910 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:46:50,913 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:46:51,012 - INFO] Load checkpoint successfully |
| [2026-06-14 08:46:51,012 - INFO] Begin to generate per chunk |
| [2026-06-14 08:46:51,013 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 08:46:51,037 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:34<06:17, 94.49s/it]
InferBatch 0: 40%|████ | 2/5 [03:08<04:43, 94.35s/it]
InferBatch 0: 60%|██████ | 3/5 [04:03<02:32, 76.46s/it]
InferBatch 0: 80%|████████ | 4/5 [04:48<01:03, 63.74s/it]
InferBatch 0: 100%|██████████| 5/5 [05:12<00:00, 49.60s/it]
InferBatch 0: 100%|██████████| 5/5 [05:14<00:00, 62.85s/it] |
| [2026-06-14 08:52:07,318 - INFO] Finish MagiPipeline, max memory allocated: 23.36 GB, max memory reserved: 24.88 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_max_w5_da0.5_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.36 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.15 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.2159dB reuse=13.44% time=351s |
| ========== [dev4] tau0.012_max_w3_da0.4_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 08:52:22.037348186 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 08:52:22,490 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 08:52:22,490 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 08:52:22,491 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.4, 'detail_lambda': 0.5, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'max'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.4 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = max |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:08<00:08, 8.88s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:17<00:00, 8.77s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:17<00:00, 8.78s/it] |
| [2026-06-14 08:52:42,674 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 08:52:42,674 - INFO] Build DiTModel successfully |
| [2026-06-14 08:52:42,674 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 08:52:42,675 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 120.40it/s] |
| [2026-06-14 08:52:44,724 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 08:52:44,725 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 08:52:44,861 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:52:44,863 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:52:44,974 - INFO] Load checkpoint successfully |
| [2026-06-14 08:52:44,974 - INFO] Begin to generate per chunk |
| [2026-06-14 08:52:44,974 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 08:52:45,042 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:36<06:26, 96.53s/it]
InferBatch 0: 40%|████ | 2/5 [03:14<04:51, 97.24s/it]
InferBatch 0: 60%|██████ | 3/5 [04:10<02:36, 78.42s/it]
InferBatch 0: 80%|████████ | 4/5 [04:54<01:04, 64.93s/it]
InferBatch 0: 100%|██████████| 5/5 [05:19<00:00, 50.37s/it]
InferBatch 0: 100%|██████████| 5/5 [05:20<00:00, 64.13s/it] |
| [2026-06-14 08:58:07,800 - INFO] Finish MagiPipeline, max memory allocated: 23.36 GB, max memory reserved: 24.92 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_max_w3_da0.4_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.36 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.25 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.2253dB reuse=13.44% time=357s |
| ========== [dev4] tau0.012_max_w3_da0.6_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 08:58:23.010582423 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 08:58:23,464 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 08:58:23,464 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 08:58:23,464 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.6, 'detail_lambda': 0.5, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'max'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.6 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = max |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.67s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.38s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.42s/it] |
| [2026-06-14 08:58:44,810 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 08:58:44,810 - INFO] Build DiTModel successfully |
| [2026-06-14 08:58:44,811 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 08:58:44,811 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 132.83it/s] |
| [2026-06-14 08:58:46,937 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 08:58:46,938 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 08:58:47,054 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:58:47,056 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 08:58:47,158 - INFO] Load checkpoint successfully |
| [2026-06-14 08:58:47,158 - INFO] Begin to generate per chunk |
| [2026-06-14 08:58:47,158 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 08:58:47,182 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:35<06:20, 95.01s/it]
InferBatch 0: 40%|████ | 2/5 [03:08<04:43, 94.38s/it]
InferBatch 0: 60%|██████ | 3/5 [04:03<02:32, 76.38s/it]
InferBatch 0: 80%|████████ | 4/5 [04:48<01:03, 63.83s/it]
InferBatch 0: 100%|██████████| 5/5 [05:13<00:00, 49.77s/it]
InferBatch 0: 100%|██████████| 5/5 [05:15<00:00, 63.01s/it] |
| [2026-06-14 09:04:04,327 - INFO] Finish MagiPipeline, max memory allocated: 23.40 GB, max memory reserved: 24.71 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_max_w3_da0.6_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.40 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.19 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.6% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.2007dB reuse=14.06% time=350s |
| ========== [dev4] tau0.012_blend_w3_da0.5_lam0.3 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 09:04:18.517162155 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 09:04:18,970 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 09:04:18,970 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 09:04:18,970 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.3, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'blend'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.3 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = blend |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.28s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.20s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.21s/it] |
| [2026-06-14 09:04:39,513 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 09:04:39,514 - INFO] Build DiTModel successfully |
| [2026-06-14 09:04:39,514 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 09:04:39,514 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 156.08it/s] |
| [2026-06-14 09:04:41,547 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 09:04:41,547 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 09:04:41,687 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:04:41,690 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:04:41,812 - INFO] Load checkpoint successfully |
| [2026-06-14 09:04:41,812 - INFO] Begin to generate per chunk |
| [2026-06-14 09:04:41,812 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 09:04:41,900 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:34<06:18, 94.54s/it]
InferBatch 0: 40%|████ | 2/5 [03:08<04:42, 94.07s/it]
InferBatch 0: 60%|██████ | 3/5 [04:04<02:33, 76.70s/it]
InferBatch 0: 80%|████████ | 4/5 [04:48<01:03, 63.90s/it]
InferBatch 0: 100%|██████████| 5/5 [05:13<00:00, 49.86s/it]
InferBatch 0: 100%|██████████| 5/5 [05:15<00:00, 63.09s/it] |
| [2026-06-14 09:09:59,557 - INFO] Finish MagiPipeline, max memory allocated: 23.48 GB, max memory reserved: 24.87 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_blend_w3_da0.5_lam0.3/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.48 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.25 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.7% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.2159dB reuse=14.06% time=351s |
| ========== [dev4] tau0.012_blend_w3_da0.5_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 09:10:15.886595567 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 09:10:15,340 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 09:10:15,340 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 09:10:15,340 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.5, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'blend'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = blend |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.89s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:19<00:00, 9.44s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:19<00:00, 9.51s/it] |
| [2026-06-14 09:10:36,364 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 09:10:36,364 - INFO] Build DiTModel successfully |
| [2026-06-14 09:10:36,364 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 09:10:36,365 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 137.13it/s] |
| [2026-06-14 09:10:38,467 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 09:10:38,467 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 09:10:38,584 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:10:38,587 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:10:38,694 - INFO] Load checkpoint successfully |
| [2026-06-14 09:10:38,694 - INFO] Begin to generate per chunk |
| [2026-06-14 09:10:38,694 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 09:10:38,719 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:32<06:10, 92.70s/it]
InferBatch 0: 40%|████ | 2/5 [03:06<04:40, 93.63s/it]
InferBatch 0: 60%|██████ | 3/5 [04:01<02:31, 75.92s/it]
InferBatch 0: 80%|████████ | 4/5 [04:46<01:03, 63.65s/it]
InferBatch 0: 100%|██████████| 5/5 [05:11<00:00, 49.54s/it]
InferBatch 0: 100%|██████████| 5/5 [05:13<00:00, 62.63s/it] |
| [2026-06-14 09:15:53,903 - INFO] Finish MagiPipeline, max memory allocated: 23.47 GB, max memory reserved: 24.73 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_blend_w3_da0.5_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.47 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.13 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.7% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.1501dB reuse=14.06% time=348s |
| ========== [dev4] tau0.012_blend_w3_da0.5_lam0.7 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 09:16:08.109278923 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 09:16:08,563 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 09:16:08,563 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 09:16:08,563 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.7, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'blend'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.7 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = blend |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.43s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.48s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:18<00:00, 9.47s/it] |
| [2026-06-14 09:16:29,511 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 09:16:29,511 - INFO] Build DiTModel successfully |
| [2026-06-14 09:16:29,512 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 09:16:29,512 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 139.98it/s] |
| [2026-06-14 09:16:31,464 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 09:16:31,464 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 09:16:31,605 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:16:31,608 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:16:31,718 - INFO] Load checkpoint successfully |
| [2026-06-14 09:16:31,719 - INFO] Begin to generate per chunk |
| [2026-06-14 09:16:31,719 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 09:16:31,757 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:30<06:02, 90.73s/it]
InferBatch 0: 40%|████ | 2/5 [03:09<04:45, 95.26s/it]
InferBatch 0: 60%|██████ | 3/5 [04:04<02:34, 77.25s/it]
InferBatch 0: 80%|████████ | 4/5 [04:48<01:04, 64.03s/it]
InferBatch 0: 100%|██████████| 5/5 [05:13<00:00, 49.91s/it]
InferBatch 0: 100%|██████████| 5/5 [05:15<00:00, 63.04s/it] |
| [2026-06-14 09:21:49,543 - INFO] Finish MagiPipeline, max memory allocated: 23.36 GB, max memory reserved: 24.94 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_blend_w3_da0.5_lam0.7/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.36 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.31 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.1782dB reuse=14.06% time=351s |
| ========== [dev4] tau0.012_product_w3_da0.5_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 09:22:05.818444946 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 09:22:05,271 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 09:22:05,271 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 09:22:05,271 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.5, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'product'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = product |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:09<00:09, 9.59s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:19<00:00, 9.96s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:19<00:00, 9.90s/it] |
| [2026-06-14 09:22:27,016 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 09:22:27,017 - INFO] Build DiTModel successfully |
| [2026-06-14 09:22:27,017 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 09:22:27,017 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 135.55it/s] |
| [2026-06-14 09:22:29,350 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 09:22:29,350 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 09:22:29,493 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:22:29,497 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:22:29,619 - INFO] Load checkpoint successfully |
| [2026-06-14 09:22:29,619 - INFO] Begin to generate per chunk |
| [2026-06-14 09:22:29,619 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 09:22:29,642 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:25<05:40, 85.10s/it]
InferBatch 0: 40%|████ | 2/5 [02:54<04:23, 87.74s/it]
InferBatch 0: 60%|██████ | 3/5 [03:45<02:21, 70.94s/it]
InferBatch 0: 80%|████████ | 4/5 [04:28<00:59, 59.85s/it]
InferBatch 0: 100%|██████████| 5/5 [04:53<00:00, 47.24s/it]
InferBatch 0: 100%|██████████| 5/5 [04:54<00:00, 58.98s/it] |
| [2026-06-14 09:27:26,687 - INFO] Finish MagiPipeline, max memory allocated: 23.35 GB, max memory reserved: 24.62 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_product_w3_da0.5_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.35 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.17 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.1277dB reuse=14.69% time=333s |
| ========== [dev4] tau0.012_product_w5_da0.5_lam0.5 (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 09:27:41.703164220 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 09:27:41,156 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 09:27:41,156 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=120, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 09:27:41,156 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.5, 'detail_window_size': 5, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'product'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.5 |
| Added to args: detail_window_size = 5 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = product |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:11<00:11, 11.71s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:22<00:00, 11.14s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:22<00:00, 11.23s/it] |
| [2026-06-14 09:28:05,819 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 09:28:05,819 - INFO] Build DiTModel successfully |
| [2026-06-14 09:28:05,819 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 09:28:05,819 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 151.46it/s] |
| [2026-06-14 09:28:08,356 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 09:28:08,356 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 09:28:08,501 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:28:08,504 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:28:08,625 - INFO] Load checkpoint successfully |
| [2026-06-14 09:28:08,625 - INFO] Begin to generate per chunk |
| [2026-06-14 09:28:08,625 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 09:28:08,648 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:24<05:39, 84.87s/it]
InferBatch 0: 40%|████ | 2/5 [02:55<04:24, 88.08s/it]
InferBatch 0: 60%|██████ | 3/5 [03:44<02:21, 70.56s/it]
InferBatch 0: 80%|████████ | 4/5 [04:27<00:59, 59.66s/it]
InferBatch 0: 100%|██████████| 5/5 [04:52<00:00, 47.05s/it]
InferBatch 0: 100%|██████████| 5/5 [04:54<00:00, 58.82s/it] |
| [2026-06-14 09:33:04,894 - INFO] Finish MagiPipeline, max memory allocated: 23.35 GB, max memory reserved: 24.65 GB |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_tau0.012_product_w5_da0.5_lam0.5/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.35 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.23 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=15.1003dB reuse=14.37% time=334s |
| ========== [dev4_full] tau0.012_blend_w3_da0.5_lam0.3_240f (PYTHONPATH=/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail:/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion) ========== |
| /home/dyvm6xra/dyvm6xrauser11/miniforge3/envs/magi/lib/python3.10/site-packages/timm/models/layers/__init__.py:48: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers |
| warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning) |
| [W614 09:33:19.999379527 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 09:33:19,452 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 09:33:19,452 - INFO] MagiConfig(model_config=ModelConfig(model_name='videodit_ardf', num_layers=34, hidden_size=3072, ffn_hidden_size=12288, num_attention_heads=24, num_query_groups=8, kv_channels=128, layernorm_epsilon=1e-06, apply_layernorm_1p=True, x_rescale_factor=1, half_channel_vae=False, params_dtype=torch.bfloat16, patch_size=2, t_patch_size=1, in_channels=16, out_channels=16, cond_hidden_ratio=0.25, caption_channels=4096, caption_max_length=800, xattn_cond_hidden_ratio=1.0, cond_gating_ratio=1.0, gated_linear_unit=False), runtime_config=RuntimeConfig(cfg_number=1, cfg_t_range=[0.0, 0.0217, 0.1, 0.3, 0.999], prev_chunk_scales=[1.5, 1.5, 1.5, 1.0, 1.0], text_scales=[7.5, 7.5, 7.5, 0.0, 0.0], noise2clean_kvrange=[], clean_chunk_kvrange=1, clean_t=0.9999, seed=1234, num_frames=240, video_size_h=720, video_size_w=720, num_steps=64, window_size=4, fps=24, chunk_width=6, t5_pretrained='./downloads/t5_pretrained', t5_device='cuda', vae_pretrained='./downloads/vae', scale_factor=0.18215, temporal_downsample_factor=4, load='./downloads/4.5B_distill'), engine_config=EngineConfig(distributed_backend='nccl', distributed_timeout_minutes=15, pp_size=1, cp_size=1, cp_strategy='none', ulysses_overlap_degree=1, fp8_quant=False, distill_nearly_clean_chunk_threshold=0.3, shortcut_mode='8,16,16', distill=True, kv_offload=False, enable_cuda_graph=False)) |
| [2026-06-14 09:33:19,452 - INFO] Precompute validation prompt embeddings |
| You are using the default legacy behaviour of the <class 'transformers.models.t5.tokenization_t5.T5Tokenizer'>. This is expected, and simply means that the `legacy` (previous) behavior will be used so nothing changes for you. If you want to use the new behaviour, set `legacy=False`. This should only be set if you understand what it means, and thoroughly read the reason why this was added as explained in https://github.com/huggingface/transformers/pull/24565 |
| Loading additional config: {'alpha': 0.5, 'compress_kv_cache': True, 'compress_strategy': 'token', 'detail_alpha': 0.5, 'detail_lambda': 0.3, 'detail_window_size': 3, 'discard_nearly_clean_chunk': True, 'log': False, 'mix_lambda': 0.07, 'phase1_steps': 9, 'power': 3, 'print_peak_memory': True, 'query_granularity': 'frame', 'rel_l1_thresh': 0.012, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5, 'weight_combine_mode': 'blend'} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| Added to args: detail_alpha = 0.5 |
| Added to args: detail_lambda = 0.3 |
| Added to args: detail_window_size = 3 |
| Added to args: discard_nearly_clean_chunk = True |
| Added to args: log = False |
| Added to args: mix_lambda = 0.07 |
| Added to args: phase1_steps = 9 |
| Added to args: power = 3 |
| Added to args: print_peak_memory = True |
| Added to args: query_granularity = frame |
| Added to args: rel_l1_thresh = 0.012 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| Added to args: weight_combine_mode = blend |
| Running on GPU: NVIDIA H800 |
| GPU Memory before pipeline: 0.00 GB |
|
Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading checkpoint shards: 50%|█████ | 1/2 [00:10<00:10, 10.57s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:20<00:00, 10.47s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:20<00:00, 10.49s/it] |
| [2026-06-14 09:33:42,983 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 09:33:42,984 - INFO] Build DiTModel successfully |
| [2026-06-14 09:33:42,984 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 09:33:42,984 - INFO] load inference_weight.distill weight from ./downloads/4.5B_distill/inference_weight.distill |
|
Loading shards: 0%| | 0/2 [00:00<?, ?it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 163.98it/s] |
| [2026-06-14 09:33:45,098 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 09:33:45,098 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 09:33:45,241 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:33:45,243 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 09:33:45,348 - INFO] Load checkpoint successfully |
| [2026-06-14 09:33:45,348 - INFO] Begin to generate per chunk |
| [2026-06-14 09:33:45,348 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/10 [00:00<?, ?it/s][2026-06-14 09:33:45,438 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 10%|█ | 1/10 [01:35<14:19, 95.54s/it]
InferBatch 0: 20%|██ | 2/10 [03:10<12:40, 95.02s/it]
InferBatch 0: 30%|███ | 3/10 [04:08<09:08, 78.29s/it]
InferBatch 0: 40%|████ | 4/10 [05:06<07:02, 70.41s/it]
InferBatch 0: 50%|█████ | 5/10 [06:05<05:31, 66.23s/it]
InferBatch 0: 60%|██████ | 6/10 [07:03<04:13, 63.32s/it]
InferBatch 0: 70%|███████ | 7/10 [08:00<03:04, 61.40s/it]
InferBatch 0: 80%|████████ | 8/10 [08:56<01:59, 59.62s/it]
InferBatch 0: 90%|█████████ | 9/10 [09:41<00:54, 54.92s/it]
InferBatch 0: 100%|██████████| 10/10 [10:05<00:00, 45.54s/it]
InferBatch 0: 100%|██████████| 10/10 [10:07<00:00, 60.75s/it] |
| [2026-06-14 09:43:56,145 - INFO] Finish MagiPipeline, max memory allocated: 23.83 GB, max memory reserved: 25.57 GB |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Using no weighting method |
| Saved MotionDetailCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev4_full_tau0.012_blend_w3_da0.5_lam0.3_240f/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.83 GB |
| Current memory allocated: 19.25 GB |
| Cached memory reserved: 20.94 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 30.1% |
| ================================================== |
| Memory after cache cleanup: 0.23 GB |
| PSNR=21.2360dB reuse=13.75% time=647s |
| Report written to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/report/comparison_report.md |
| Wrote best config yaml files |
| Done. Report: /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/report/comparison_report.md |
|
|
