| Sweep output: /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749 (num_frames=120, GPU=1, host=hk01dgx013) |
|
|
| ========== [dev3] tau0.010 ========== |
| /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 06:38:01.303500843 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 06:38:01,759 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 06:38:01,759 - 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 06:38:01,759 - 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', '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': 0.01, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.01 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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 [00:46<00:46, 46.30s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [01:29<00:00, 44.36s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [01:29<00:00, 44.65s/it] |
| [2026-06-14 06:39:34,300 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 06:39:34,300 - INFO] Build DiTModel successfully |
| [2026-06-14 06:39:34,300 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 06:39:34,300 - 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: 50%|█████ | 1/2 [00:00<00:00, 1.75it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 2.06it/s]
Loading shards: 100%|██████████| 2/2 [00:00<00:00, 2.01it/s] |
| [2026-06-14 06:39:56,740 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 06:39:56,740 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 06:39:56,910 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:39:56,912 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:39:57,043 - INFO] Load checkpoint successfully |
| [2026-06-14 06:39:57,043 - INFO] Begin to generate per chunk |
| [2026-06-14 06:39:57,043 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 06:39:57,246 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:43<06:54, 103.66s/it]
InferBatch 0: 40%|████ | 2/5 [03:52<05:55, 118.36s/it]
InferBatch 0: 60%|██████ | 3/5 [04:49<03:00, 90.31s/it]
InferBatch 0: 80%|████████ | 4/5 [05:33<01:12, 72.23s/it]
InferBatch 0: 100%|██████████| 5/5 [05:58<00:00, 55.01s/it]
InferBatch 0: 100%|██████████| 5/5 [05:59<00:00, 71.95s/it] |
| [2026-06-14 06:45:59,470 - INFO] Finish MagiPipeline, max memory allocated: 23.40 GB, max memory reserved: 25.03 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.010/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.40 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.32 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.6% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=NAdB reuse=NA% time=492s |
|
|
| ========== [dev3] tau0.012 ========== |
| /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 06:46:11.783949988 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 06:46:11,237 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 06:46:11,237 - 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 06:46:11,237 - 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', '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} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 |
| 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, 8.68s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:17<00:00, 9.04s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:17<00:00, 8.99s/it] |
| [2026-06-14 06:46:30,923 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 06:46:30,923 - INFO] Build DiTModel successfully |
| [2026-06-14 06:46:30,923 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 06:46:30,923 - 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, 157.11it/s] |
| [2026-06-14 06:46:32,700 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 06:46:32,700 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 06:46:33,058 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:46:33,061 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:46:33,181 - INFO] Load checkpoint successfully |
| [2026-06-14 06:46:33,181 - INFO] Begin to generate per chunk |
| [2026-06-14 06:46:33,182 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 06:46:33,206 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:35<06:23, 95.89s/it]
InferBatch 0: 40%|████ | 2/5 [03:11<04:47, 95.70s/it]
InferBatch 0: 60%|██████ | 3/5 [04:07<02:35, 77.62s/it]
InferBatch 0: 80%|████████ | 4/5 [04:51<01:04, 64.43s/it]
InferBatch 0: 100%|██████████| 5/5 [05:16<00:00, 50.01s/it]
InferBatch 0: 100%|██████████| 5/5 [05:17<00:00, 63.55s/it] |
| [2026-06-14 06:51:53,067 - INFO] Finish MagiPipeline, max memory allocated: 23.36 GB, max memory reserved: 24.74 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.012/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.36 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.11 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=NAdB reuse=NA% time=350s |
|
|
| ========== [dev3] tau0.015 ========== |
| /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 06:52:03.168909328 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 06:52:03,621 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 06:52:03,621 - 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 06:52:03,621 - 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', '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': 0.015, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.015 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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, 8.29s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.44s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.42s/it] |
| [2026-06-14 06:52:22,315 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 06:52:22,315 - INFO] Build DiTModel successfully |
| [2026-06-14 06:52:22,315 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 06:52:22,315 - 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, 134.99it/s] |
| [2026-06-14 06:52:23,882 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 06:52:23,882 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 06:52:24,041 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:52:24,044 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:52:24,150 - INFO] Load checkpoint successfully |
| [2026-06-14 06:52:24,150 - INFO] Begin to generate per chunk |
| [2026-06-14 06:52:24,150 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 06:52:24,179 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:27<05:50, 87.71s/it]
InferBatch 0: 40%|████ | 2/5 [02:56<04:25, 88.56s/it]
InferBatch 0: 60%|██████ | 3/5 [03:48<02:23, 71.73s/it]
InferBatch 0: 80%|████████ | 4/5 [04:31<01:00, 60.47s/it]
InferBatch 0: 100%|██████████| 5/5 [04:56<00:00, 47.52s/it]
InferBatch 0: 100%|██████████| 5/5 [04:57<00:00, 59.58s/it] |
| [2026-06-14 06:57:24,139 - INFO] Finish MagiPipeline, max memory allocated: 23.35 GB, max memory reserved: 24.84 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.015/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=NAdB reuse=NA% time=329s |
|
|
| ========== [dev3] tau0.018 ========== |
| /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 06:57:34.299740065 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 06:57:34,753 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 06:57:34,753 - 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 06:57:34,753 - 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', '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': 0.018, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.018 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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, 8.30s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.32s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.32s/it] |
| [2026-06-14 06:57:53,084 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 06:57:53,084 - INFO] Build DiTModel successfully |
| [2026-06-14 06:57:53,084 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 06:57:53,084 - 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, 141.63it/s] |
| [2026-06-14 06:57:54,710 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 06:57:54,710 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 06:57:54,850 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:57:54,853 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 06:57:54,954 - INFO] Load checkpoint successfully |
| [2026-06-14 06:57:54,954 - INFO] Begin to generate per chunk |
| [2026-06-14 06:57:54,954 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 06:57:54,977 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:22<05:30, 82.56s/it]
InferBatch 0: 40%|████ | 2/5 [02:49<04:15, 85.31s/it]
InferBatch 0: 60%|██████ | 3/5 [03:38<02:17, 68.79s/it]
InferBatch 0: 80%|████████ | 4/5 [04:21<00:58, 58.58s/it]
InferBatch 0: 100%|██████████| 5/5 [04:46<00:00, 46.37s/it]
InferBatch 0: 100%|██████████| 5/5 [04:48<00:00, 57.61s/it] |
| [2026-06-14 07:02:45,116 - INFO] Finish MagiPipeline, max memory allocated: 23.35 GB, max memory reserved: 24.94 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.018/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=NAdB reuse=NA% time=320s |
|
|
| ========== [dev3] tau0.020 ========== |
| /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 07:02:56.745687470 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 07:02:56,199 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 07:02:56,199 - 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 07:02:56,199 - 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', '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': 0.02, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.02 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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 [00:07<00:07, 7.91s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:15<00:00, 7.98s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:15<00:00, 7.97s/it] |
| [2026-06-14 07:03:13,960 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 07:03:13,960 - INFO] Build DiTModel successfully |
| [2026-06-14 07:03:13,960 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 07:03:13,960 - 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, 151.99it/s] |
| [2026-06-14 07:03:15,874 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 07:03:15,875 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 07:03:16,013 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:03:16,016 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:03:16,131 - INFO] Load checkpoint successfully |
| [2026-06-14 07:03:16,131 - INFO] Begin to generate per chunk |
| [2026-06-14 07:03:16,131 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 07:03:16,155 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:20<05:20, 80.05s/it]
InferBatch 0: 40%|████ | 2/5 [02:45<04:10, 83.44s/it]
InferBatch 0: 60%|██████ | 3/5 [03:33<02:13, 66.99s/it]
InferBatch 0: 80%|████████ | 4/5 [04:15<00:57, 57.16s/it]
InferBatch 0: 100%|██████████| 5/5 [04:39<00:00, 45.40s/it]
InferBatch 0: 100%|██████████| 5/5 [04:41<00:00, 56.29s/it] |
| [2026-06-14 07:07:59,821 - INFO] Finish MagiPipeline, max memory allocated: 23.35 GB, max memory reserved: 24.72 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.020/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.35 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.09 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.5% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=NAdB reuse=NA% time=313s |
|
|
| ========== [dev3] tau0.025 ========== |
| /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 07:08:11.894849908 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 07:08:11,348 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 07:08:11,348 - 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 07:08:11,348 - 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', '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': 0.025, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.025 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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, 8.33s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.29s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.30s/it] |
| [2026-06-14 07:08:29,636 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 07:08:29,636 - INFO] Build DiTModel successfully |
| [2026-06-14 07:08:29,636 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 07:08:29,636 - 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, 132.76it/s] |
| [2026-06-14 07:08:31,276 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 07:08:31,276 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 07:08:31,419 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:08:31,421 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:08:31,535 - INFO] Load checkpoint successfully |
| [2026-06-14 07:08:31,535 - INFO] Begin to generate per chunk |
| [2026-06-14 07:08:31,535 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 07:08:31,559 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:12<04:49, 72.46s/it]
InferBatch 0: 40%|████ | 2/5 [02:34<03:54, 78.32s/it]
InferBatch 0: 60%|██████ | 3/5 [03:17<02:03, 61.82s/it]
InferBatch 0: 80%|████████ | 4/5 [03:57<00:53, 53.46s/it]
InferBatch 0: 100%|██████████| 5/5 [04:22<00:00, 43.01s/it]
InferBatch 0: 100%|██████████| 5/5 [04:23<00:00, 52.75s/it] |
| [2026-06-14 07:12:57,458 - INFO] Finish MagiPipeline, max memory allocated: 23.34 GB, max memory reserved: 24.56 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.025/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.34 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=NAdB reuse=NA% time=296s |
|
|
| ========== [dev3] tau0.030 ========== |
| /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 07:13:08.662415941 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 07:13:08,115 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 07:13:08,115 - 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 07:13:08,115 - 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', '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': 0.03, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.03 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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, 8.37s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.15s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.18s/it] |
| [2026-06-14 07:13:26,070 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 07:13:26,070 - INFO] Build DiTModel successfully |
| [2026-06-14 07:13:26,070 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 07:13:26,070 - 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, 154.18it/s] |
| [2026-06-14 07:13:27,698 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 07:13:27,698 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 07:13:27,819 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:13:27,821 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:13:27,914 - INFO] Load checkpoint successfully |
| [2026-06-14 07:13:27,914 - INFO] Begin to generate per chunk |
| [2026-06-14 07:13:27,914 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/5 [00:00<?, ?it/s][2026-06-14 07:13:27,937 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 20%|██ | 1/5 [01:05<04:20, 65.16s/it]
InferBatch 0: 40%|████ | 2/5 [02:24<03:39, 73.23s/it]
InferBatch 0: 60%|██████ | 3/5 [03:02<01:55, 57.52s/it]
InferBatch 0: 80%|████████ | 4/5 [03:42<00:50, 50.36s/it]
InferBatch 0: 100%|██████████| 5/5 [04:06<00:00, 40.97s/it]
InferBatch 0: 100%|██████████| 5/5 [04:08<00:00, 49.62s/it] |
| [2026-06-14 07:17:38,229 - INFO] Finish MagiPipeline, max memory allocated: 23.28 GB, max memory reserved: 24.96 GB |
| Saved MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_tau0.030/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.28 GB |
| Current memory allocated: 18.98 GB |
| Cached memory reserved: 20.29 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 29.4% |
| ================================================== |
| Memory after cache cleanup: 0.16 GB |
| PSNR=NAdB reuse=NA% time=279s |
| Best dev3 tau from sweep: 0.015 |
|
|
| ========== [dev4] tau0.015_max_w3_da0.5_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_max_w3_da0.5_lam0.5 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_max_w5_da0.5_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_max_w5_da0.5_lam0.5 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_max_w3_da0.4_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_max_w3_da0.4_lam0.5 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_max_w3_da0.6_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_max_w3_da0.6_lam0.5 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_blend_w3_da0.5_lam0.3 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_blend_w3_da0.5_lam0.3 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_blend_w3_da0.5_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_blend_w3_da0.5_lam0.5 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_blend_w3_da0.5_lam0.7 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_blend_w3_da0.5_lam0.7 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_product_w3_da0.5_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_product_w3_da0.5_lam0.5 (rc=1, no video) |
|
|
| ========== [dev4] tau0.015_product_w5_da0.5_lam0.5 ========== |
| /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) |
| Traceback (most recent call last): |
| File "/home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev4-detail/inference/pipeline/motioncache.py", line 34, in <module> |
| from inference.pipeline.cache.motiondetailcache import MotionDetailCache |
| ModuleNotFoundError: No module named 'inference.pipeline.cache.motiondetailcache' |
| FAILED: tau0.015_product_w5_da0.5_lam0.5 (rc=1, no video) |
| Full validation at 240 frames... |
|
|
| ========== [dev3_full] tau0.012_240f ========== |
| /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 07:18:29.363647792 CUDAAllocatorConfig.h:28] Warning: expandable_segments not supported on this platform (function operator()) |
| [2026-06-14 07:18:29,816 - INFO] Initialize torch distribution and model parallel successfully |
| [2026-06-14 07:18:29,816 - 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 07:18:29,816 - 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', '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': 0.015, 'score_weighting_method': 'no_weight', 'total_cache_chunk_nums': 5, 'warmup_steps': 5} |
| Added to args: alpha = 0.5 |
| Added to args: compress_kv_cache = True |
| Added to args: compress_strategy = token |
| 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 = 0.015 |
| Added to args: score_weighting_method = no_weight |
| Added to args: total_cache_chunk_nums = 5 |
| Added to args: warmup_steps = 5 |
| 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, 8.39s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.43s/it]
Loading checkpoint shards: 100%|██████████| 2/2 [00:16<00:00, 8.42s/it] |
| [2026-06-14 07:18:48,218 - INFO] (cp, pp) rank (0, 0): param count 4459898128, model size 8.34 GB |
| [2026-06-14 07:18:48,218 - INFO] Build DiTModel successfully |
| [2026-06-14 07:18:48,219 - INFO] After build_dit_model, memory allocated: 0.04 GB, memory reserved: 0.08 GB |
| [2026-06-14 07:18:48,219 - 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, 141.14it/s] |
| [2026-06-14 07:18:49,778 - INFO] Load Weight Missing Keys: [] |
| [2026-06-14 07:18:49,778 - INFO] Load Weight Unexpected Keys: [] |
| [2026-06-14 07:18:49,915 - INFO] After load_checkpoint, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:18:49,918 - INFO] After high_precision_promoter, memory allocated: 8.39 GB, memory reserved: 8.40 GB |
| [2026-06-14 07:18:50,013 - INFO] Load checkpoint successfully |
| [2026-06-14 07:18:50,013 - INFO] Begin to generate per chunk |
| [2026-06-14 07:18:50,013 - INFO] special_token = ['HQ_TOKEN', 'DURATION_TOKEN'] |
|
InferBatch 0: 0%| | 0/10 [00:00<?, ?it/s][2026-06-14 07:18:50,037 - INFO] transport_inputs len: 1 |
|
InferBatch 0: 10%|█ | 1/10 [01:29<13:22, 89.13s/it]
InferBatch 0: 20%|██ | 2/10 [03:00<12:03, 90.42s/it]
InferBatch 0: 30%|███ | 3/10 [03:55<08:39, 74.28s/it]
InferBatch 0: 40%|████ | 4/10 [04:51<06:42, 67.09s/it]
InferBatch 0: 50%|█████ | 5/10 [05:47<05:14, 62.89s/it]
InferBatch 0: 60%|██████ | 6/10 [06:42<04:00, 60.21s/it]
InferBatch 0: 70%|███████ | 7/10 [07:36<02:54, 58.18s/it]
InferBatch 0: 80%|████████ | 8/10 [08:28<01:52, 56.38s/it]
InferBatch 0: 90%|█████████ | 9/10 [09:12<00:52, 52.43s/it]
InferBatch 0: 100%|██████████| 10/10 [09:36<00:00, 43.81s/it]
InferBatch 0: 100%|██████████| 10/10 [09:38<00:00, 57.84s/it] |
| [2026-06-14 07:28:31,755 - INFO] Finish MagiPipeline, max memory allocated: 23.87 GB, max memory reserved: 28.47 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 MotionCache metric stats to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/dev3_full_tau0.012_240f/metrics.json |
| ✅ Video saved successfully. |
|
|
| ================================================== |
| GPU Memory Usage Summary: |
| Peak memory allocated: 23.87 GB |
| Current memory allocated: 19.25 GB |
| Cached memory reserved: 21.07 GB |
| Total GPU memory: 79.11 GB |
| Peak memory usage: 30.2% |
| ================================================== |
| Memory after cache cleanup: 0.23 GB |
| PSNR=20.4377dB reuse=15.62% time=609s |
| Report written to /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/report/comparison_report.md |
|
|
| Sweep complete. |
| CSV: /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/report/results.csv |
| Report: /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1-dev3-motion/outputs/hparam_sweep_20260614_063749/report/comparison_report.md |
|
|
