SnoopyFan commited on
Commit
d6c49cd
·
verified ·
1 Parent(s): 0121268

Add WorldCompass Sana training artifacts

Browse files
rebuttal_data/worldcompass_sana/wc_sana_smoke_027c9c8_20260802f/node_0/torchrun.log ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /mnt/workspace/fly/reward_dmd_sanawm/.venv/lib/python3.11/site-packages/mmcv/__init__.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details.
2
+ warnings.warn(
3
+ /mnt/workspace/fly/reward_dmd_sanawm/.venv/lib/python3.11/site-packages/mmcv/__init__.py:20: UserWarning: On January 1, 2023, MMCV will release v2.0.0, in which it will remove components related to the training process and add a data transformation module. In addition, it will rename the package names mmcv to mmcv-lite and mmcv-full to mmcv. See https://github.com/open-mmlab/mmcv/blob/master/docs/en/compatibility.md for more details.
4
+ warnings.warn(
5
+ /mnt/workspace/fly/reward_dmd_sanawm/Sana_RF_24card/diffusion/model/wan/model.py:124: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
6
+ @amp.autocast(enabled=False)
7
+ /mnt/workspace/fly/reward_dmd_sanawm/Sana_RF_24card/diffusion/model/wan/model.py:124: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
8
+ @amp.autocast(enabled=False)
9
+ /mnt/workspace/fly/reward_dmd_sanawm/Sana_RF_24card/diffusion/model/wan/model.py:134: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
10
+ @amp.autocast(enabled=False)
11
+ /mnt/workspace/fly/reward_dmd_sanawm/Sana_RF_24card/diffusion/model/wan/model.py:134: FutureWarning: `torch.cuda.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cuda', args...)` instead.
12
+ @amp.autocast(enabled=False)
13
+ /mnt/workspace/fly/reward_dmd_sanawm/Sana_RF_24card/diffusion/model/nets/fastlinear/modules/nn/norm.py:186: UserWarning: Cannot import apex RMSNorm, switch to vanilla implementation
14
+ warnings.warn("Cannot import apex RMSNorm, switch to vanilla implementation")
15
+ /mnt/workspace/fly/reward_dmd_sanawm/Sana_RF_24card/diffusion/model/nets/fastlinear/modules/nn/norm.py:186: UserWarning: Cannot import apex RMSNorm, switch to vanilla implementation
16
+ warnings.warn("Cannot import apex RMSNorm, switch to vanilla implementation")
17
+ 2026-08-02 01:38:37 - [Sana] - INFO - [FSDP2] block=SanaVideoMSCamCtrlBlock, cp_size=1, reshard_after_forward=True, limit_all_gathers=False, backward_prefetch=backward_pre
18
+ backward_prefetch is not supported in FSDP2. Setting backward prefetch to None.
19
+ backward_prefetch is not supported in FSDP2. Setting backward prefetch to None.
20
+ 2026-08-02 01:38:41 - [Sana] - INFO - Failed to load null embed from None....'NoneType' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.. Ignore the error during inference
21
+ 2026-08-02 01:38:41 - [Sana] - INFO - use pe: True, pos embed type: casual_wan_rope, position embed interpolation: 1.0, base size: 22
22
+ 2026-08-02 01:38:41 - [Sana] - INFO - attention type: BidirectionalGDNTriton; ffn type: CachedGLUMBConvTemp; self-attn head dim: 112; self-attn qk norm: True; cross-attn type: flash; cross-attn qk norm: True; autocast linear attn: False
23
+ 2026-08-02 01:38:41 - [Sana] - INFO - Hybrid attention (softmax_every_n=4):
24
+ attn_type_list = ['BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton', 'BidirectionalGDNTriton']
25
+ camctrl_cls_list = ['CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedSoftmaxUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedSoftmaxUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedSoftmaxUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedSoftmaxUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedChunkCausalGDNUCPESinglePathLiteLA', 'CachedSoftmaxUCPESinglePathLiteLA']
26
+ 2026-08-02 01:38:59 - [Sana] - INFO - Failed to load null embed from None....'NoneType' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.. Ignore the error during inference
27
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
28
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
29
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
30
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
31
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
32
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
33
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
34
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
35
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
36
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
37
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
38
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
39
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
40
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
41
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
42
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
43
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
44
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
45
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
46
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
47
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
48
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
49
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
50
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
51
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
52
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
53
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
54
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
55
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
56
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
57
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
58
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
59
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
60
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
61
+ [Sana] Loading model from /mnt/workspace/fly/reward_dmd_sanawm/models/SANA-WM_streaming/fb4ef609f5da47d5815e850a5ecec9a12f069b57/sana_dit/model.pt
62
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
63
+ [Sana] Loaded model from /mnt/workspace/fly/reward_dmd_sanawm/models/SANA-WM_streaming/fb4ef609f5da47d5815e850a5ecec9a12f069b57/sana_dit/model.pt
64
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
65
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
66
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
67
+ Initializing camera branch QKV from base model QKV for CachedChunkCausalGDNUCPESinglePathLiteLA
68
+ Initializing camera branch QKV from base model QKV for CachedSoftmaxUCPESinglePathLiteLA
69
+ [Sana] Loading model from /mnt/workspace/fly/reward_dmd_sanawm/models/SANA-WM_streaming/fb4ef609f5da47d5815e850a5ecec9a12f069b57/sana_dit/model.pt
70
+ [Sana] Loaded model from /mnt/workspace/fly/reward_dmd_sanawm/models/SANA-WM_streaming/fb4ef609f5da47d5815e850a5ecec9a12f069b57/sana_dit/model.pt
71
+ 2026-08-02 01:39:48 - [Sana] - INFO - Loaded /mnt/workspace/fly/reward_dmd_sanawm/models/SANA-WM_streaming/fb4ef609f5da47d5815e850a5ecec9a12f069b57/sana_dit/model.pt: model=SanaMSVideoCamCtrlStreaming_1600M_P1_D20, missing=1, unexpected=0
72
+ 2026-08-02 01:39:51 - [Sana] - INFO - FSDP2: removing custom __call__ from SanaMSVideoCamCtrlStreaming
73
+ 2026-08-02 01:39:51 - [Sana] - INFO - FSDP2: removing custom __call__ from Sana
74
+ 2026-08-02 01:39:51 - [Sana] - INFO - FSDP2: removing custom __call__ from SanaMSVideoCamCtrlStreaming
75
+ 2026-08-02 01:39:51 - [Sana] - INFO - FSDP2: removing custom __call__ from Sana
76
+
77
+
78
+ 2026-08-02 01:39:57 - [Sana] - INFO - Constructing dataset ThreeSourceF19ManifestDataset...
79
+ 2026-08-02 01:40:15 - [Sana] - INFO - Dataset ThreeSourceF19ManifestDataset constructed: time: 18.18 s, length (use/ori): 112573/112573
80
+ NCCL version 2.27.5+cuda12.9
81
+ 2026-08-02 01:40:23 - [Sana] - INFO - WorldCompass released-code SANA port: profile=released_code_1588e133_sana_native_smoke G=6 top=3 bottom=3 target_optimizer_steps=3 target_outer_steps=1 N=6 schedule=[1000,960,889,727,0] full_dit=true optimizer=Muon(lr=1e-5,wd=1e-4,momentum=0.95) reward_decode=stage1_causal_vae_before_frozen_refiner evaluation_checkpoint=model_ema.pt
82
+ 2026-08-02 01:40:23 - [Sana] - INFO - mode=worldcompass start_optimizer_step=0 target_optimizer_steps=3 start_outer_step=0 target_outer_steps=1 optimizer_updates_per_outer_step=3 save_every_optimizer_steps=3 save_enabled=True cp_size=1 dp_size=2 reward_domain=stage1_causal_vae_before_frozen_refiner
83
+ 2026-08-02 01:42:05 - [Sana] - INFO - mode=worldcompass optimizer_step=3 outer_step=1 action_advantage_mean=0.000000 action_joint_accuracy=0.680556 action_rotation_accuracy=0.743056 action_translation_accuracy=0.913194 action_translation_threshold=0.002000 backward_calls=12.000000 candidate_count=6.000000 checkpoint_ema_decay=0.900000 combined_advantage_max=0.334194 combined_advantage_min=-0.326783 current_policy_forwards=12.000000 ema_updates=3.000000 global_action_std=0.122166 global_first_frame_quality=7.157005 global_visual_std=1.887625 negative_loss_mean=0.833659 old_policy_forwards=12.000000 optimality_probability_max=0.583548 optimality_probability_min=0.418304 optimizer_step_after_outer=3.000000 optimizer_updates=3.000000 outer_seconds=101.665108 outer_step_after_transaction=1.000000 policy_ema_decay=0.210000 policy_final_loss_sum=4.996094 policy_grad_norm_max=4.250000 policy_loss_mean=0.832682 positive_loss_mean=0.831706 raw_action_mean=0.828125 raw_visual_mean=-2.751762 reward_da3_calls=6.000000 reward_da3_seconds=6.179152 reward_decode_calls=7.000000 reward_decode_seconds=8.845667 reward_hps_calls=7.000000 reward_hps_seconds=6.785281 reward_wall_seconds=24.554188 rollout_denoise_forwards=24.000000 selected_candidate_count=6.000000 target_chunk=0.000000 visual_advantage_mean=-0.000000 optimizer_step=3.000000 outer_step=1.000000 target_optimizer_steps=3.000000 target_outer_steps=1.000000 remaining_optimizer_steps=0.000000
84
+ 2026-08-02 01:42:44 - [Sana] - INFO - mode=worldcompass checkpoint_committed=true optimizer_step=3 outer_step=1 path=/mnt/workspace/fly/reward_dmd_sanawm/output/worldcompass_sana/wc_sana_smoke_027c9c8_20260802f/checkpoints/optimizer_step_000003_outer_step_000001
85
+ 2026-08-02 01:42:44 - [Sana] - INFO - mode=worldcompass training_complete=true optimizer_step=3 outer_step=1 checkpoint_optimizer_step=3
86
+ [W802 01:42:46.479950360 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
87
+ [W802 01:42:47.792245650 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
88
+ [W802 01:42:47.686660049 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())