Cccccz commited on
Commit
5be8366
·
verified ·
1 Parent(s): 11a7e16

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_082.json +53 -0
  2. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_083.json +53 -0
  3. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_084.json +53 -0
  4. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_085.json +53 -0
  5. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_086.json +53 -0
  6. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_087.json +53 -0
  7. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_088.json +53 -0
  8. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_089.json +53 -0
  9. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_090.json +53 -0
  10. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_091.json +53 -0
  11. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_092.json +53 -0
  12. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_000.json +53 -0
  13. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_001.json +53 -0
  14. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_002.json +53 -0
  15. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_003.json +53 -0
  16. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_004.json +53 -0
  17. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_005.json +53 -0
  18. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_006.json +53 -0
  19. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_007.json +53 -0
  20. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_008.json +53 -0
  21. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_009.json +53 -0
  22. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_010.json +53 -0
  23. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_011.json +53 -0
  24. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_012.json +53 -0
  25. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_013.json +53 -0
  26. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_014.json +53 -0
  27. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_015.json +53 -0
  28. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_016.json +53 -0
  29. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_017.json +53 -0
  30. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_018.json +53 -0
  31. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_019.json +53 -0
  32. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_020.json +53 -0
  33. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_021.json +53 -0
  34. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_022.json +53 -0
  35. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_023.json +53 -0
  36. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_024.json +53 -0
  37. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_025.json +53 -0
  38. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_026.json +53 -0
  39. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_027.json +53 -0
  40. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_028.json +53 -0
  41. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_029.json +53 -0
  42. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_030.json +53 -0
  43. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_031.json +53 -0
  44. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_032.json +53 -0
  45. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_033.json +53 -0
  46. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_034.json +53 -0
  47. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_035.json +53 -0
  48. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_036.json +53 -0
  49. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_037.json +53 -0
  50. eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_038.json +53 -0
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_082.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 765,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 82,
25
+ "prompt": "In a futuristic Tokyo rooftop set during a heavy rainstorm, a robot DJ is spinning records at a cyberpunk nightclub. The robot has sleek metallic arms and a glowing LED screen for a face, expressing intense concentration as it mixes tracks. The DJ is surrounded by neon lights and holographic displays, casting colorful reflections off the wet surroundings. The rain creates a misty atmosphere, enhancing the dystopian feel. The scene includes other robots and humans dancing under umbrellas, adding dynamic movement. Close-up shots of the DJ's hands manipulating the turntables and a wide shot capturing the lively rooftop party. Sci-fi, fantasy style.",
26
+ "original_prompt": "A robot DJ is playing the turntable, in heavy raining futuristic tokyo rooftop cyberpunk night, sci-fi, fantasy",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/082.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1216.8995552062988,
34
+ "excluded_context_kv_latency_ms": 786.7516708374023,
35
+ "wall_generation_s": 3.723658561706543,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.0378299206495285,
39
+ "psnr": 14.221645593643188,
40
+ "ssim": 0.5234529376029968,
41
+ "lpips": 0.3780893385410309,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_083.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 766,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 83,
25
+ "prompt": "A large great white shark swimming gracefully through the crystal-clear waters of the Caribbean Ocean. The shark glides smoothly with its powerful tail fin moving side to side, creating gentle ripples in the water. Its sleek, streamlined body is covered in small, shiny scales, reflecting the sunlight that penetrates the ocean depths. Schools of colorful tropical fish swim around the shark, adding vibrant life to the underwater scene. The background showcases the sandy ocean floor dotted with coral reefs and patches of seaweed swaying gently with the current. The camera follows the shark from a medium underwater perspective, capturing the shark's fluid motions and the serene beauty of its surroundings.",
26
+ "original_prompt": "A shark swimming in clear Caribbean ocean",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/083.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1222.656319618225,
34
+ "excluded_context_kv_latency_ms": 791.1024627685547,
35
+ "wall_generation_s": 3.7359237670898438,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.04499896988272667,
39
+ "psnr": 13.467974662780762,
40
+ "ssim": 0.5348247289657593,
41
+ "lpips": 0.40129366517066956,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_084.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 767,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 84,
25
+ "prompt": "A futuristic super robot standing tall and vigilant, protecting a bustling city skyline from a looming threat. The robot has a sleek, metallic design with glowing blue energy lines running across its body. It stands in a heroic pose, arms raised, ready to defend the city below. The urban landscape features towering skyscrapers, bustling streets, and neon lights reflecting off the wet pavement after a rain shower. The scene is captured in a sweeping wide shot, showcasing the robot's imposing size and the vibrant city life behind it.",
26
+ "original_prompt": "A super robot protecting city",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/084.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1225.6019096374512,
34
+ "excluded_context_kv_latency_ms": 794.6364440917969,
35
+ "wall_generation_s": 3.7415966987609863,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.04246418550610542,
39
+ "psnr": 13.719772100448608,
40
+ "ssim": 0.4770912826061249,
41
+ "lpips": 0.3061571419239044,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_085.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 768,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 85,
25
+ "prompt": "A cute, soft teddy bear with brown fur and black button eyes is washing dishes in a kitchen sink. The teddy bear stands on its hind legs, holding a sponge in one paw and a dish in the other, scrubbing away as if it were a real person doing chores. Its fluffy body and friendly expression contrast with the domestic task, creating a whimsical scene. The kitchen is simply decorated with white tiles and a few utensils on the counter. The teddy bear's posture is upright and focused, emphasizing its involvement in the activity. Medium shot focusing on the teddy bear's actions.",
26
+ "original_prompt": "A teddy bear washing the dishes",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/085.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1209.7541217803955,
34
+ "excluded_context_kv_latency_ms": 784.1963043212891,
35
+ "wall_generation_s": 3.7137529850006104,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.01993836835026741,
39
+ "psnr": 17.003103494644165,
40
+ "ssim": 0.7583621144294739,
41
+ "lpips": 0.21358801424503326,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_086.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 769,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 86,
25
+ "prompt": "In an epic scene, a massive tornado composed of swirling smoke descends from the night sky, attacking a futuristic city bathed in neon lights and glow. The tornado is dark and ominous, with tendrils of smoke reaching out to engulf towering skyscrapers and streets below. The city glows vibrantly, with colorful lights reflecting off the storm clouds and rain. Buildings are illuminated by emergency lights and flickering street lamps as the tornado twists and turns above, casting dramatic shadows. The scene captures the chaos and intensity of nature's wrath against human civilization, with a wide-angle view to showcase the scale and destruction.",
26
+ "original_prompt": "An epic tornado attacking above a glowing city at night, the tornado is made of smoke",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/086.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1223.8253479003906,
34
+ "excluded_context_kv_latency_ms": 791.2688140869141,
35
+ "wall_generation_s": 3.737698793411255,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03531907498836517,
39
+ "psnr": 14.519907236099243,
40
+ "ssim": 0.4826070964336395,
41
+ "lpips": 0.41495391726493835,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_087.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 770,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 87,
25
+ "prompt": "Oil painting style, depicting a couple dressed in elegant evening attire walking home under heavy rain. The man is wearing a black tuxedo with a bow tie, while the woman is in a flowing evening gown with a fitted bodice and full skirt, adorned with intricate lace and embroidery. They are holding umbrellas, but the rain is so intense that water droplets are visible around them. The background showcases a dimly lit city street with blurred lights from distant buildings. Both figures are positioned close together, sharing an umbrella, with a slightly hunched posture due to the rain. The scene captures the romantic yet challenging atmosphere of a sudden downpour. Medium shot, focusing on the couple's interaction and the surrounding environment.",
26
+ "original_prompt": "An oil painting of a couple in formal evening wear going home get caught in a heavy downpour with umbrellas",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/087.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1229.3607215881348,
34
+ "excluded_context_kv_latency_ms": 795.4045104980469,
35
+ "wall_generation_s": 3.746828079223633,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.037253040820360184,
39
+ "psnr": 14.288382530212402,
40
+ "ssim": 0.5409795641899109,
41
+ "lpips": 0.38980022072792053,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_088.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 771,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 88,
25
+ "prompt": "A vibrant underwater scene featuring several clownfish swimming gracefully through a colorful coral reef. The clownfish have distinctive orange bodies with white bars and black outlines, darting among the intricate and diverse coral formations. The corals are a mix of soft and hard varieties, showcasing various shapes and hues such as pink, green, and purple. The water is clear, with sunlight filtering through, creating a serene and lively atmosphere. The camera remains static, capturing the continuous motion of the clownfish as they explore their environment. Close-up view.",
26
+ "original_prompt": "Clown fish swimming through the coral reef",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/088.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1217.236961364746,
34
+ "excluded_context_kv_latency_ms": 787.2274780273438,
35
+ "wall_generation_s": 3.724541187286377,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.08412054926156998,
39
+ "psnr": 10.75097918510437,
40
+ "ssim": 0.3936209976673126,
41
+ "lpips": 0.45063716173171997,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_089.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 772,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 89,
25
+ "prompt": "A hyper-realistic depiction of a sleek, metallic spaceship descending onto the rugged surface of Mars. The spaceship, with intricate details like aerodynamic panels and glowing engines, touches down amidst swirling clouds of red dust. The Martian landscape features jagged rocks, vast plains, and distant craters under a starry sky. The camera captures the dramatic moment from a low-angle view, emphasizing the scale and detail of the spacecraft as it lands softly on the alien terrain. Wide shot, static scene.",
26
+ "original_prompt": "Hyper-realistic spaceship landing on Mars",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/089.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1219.4550218582153,
34
+ "excluded_context_kv_latency_ms": 790.5132522583008,
35
+ "wall_generation_s": 3.731562376022339,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.02982219308614731,
39
+ "psnr": 15.254604816436768,
40
+ "ssim": 0.6427234411239624,
41
+ "lpips": 0.3099578022956848,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_090.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 773,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 90,
25
+ "prompt": "A bustling scene from The Bund in Shanghai, featuring vibrant colors and lively activity. The video showcases a mix of traditional and modern architecture, including historic buildings and contemporary skyscrapers. Brightly lit storefronts, colorful advertisements, and energetic crowds fill the streets. The Huangpu River flows in the background, reflecting the city lights. People walk briskly, shop at street vendors, and take in the scenic views. The camera captures the dynamic energy of the area in a wide shot, emphasizing the blend of cultures and the vibrant atmosphere.",
26
+ "original_prompt": "The bund Shanghai, vibrant color",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/090.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1228.479287147522,
34
+ "excluded_context_kv_latency_ms": 794.6130599975586,
35
+ "wall_generation_s": 3.7458064556121826,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.04182593151926994,
39
+ "psnr": 13.785543441772461,
40
+ "ssim": 0.5153733491897583,
41
+ "lpips": 0.2942996621131897,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_091.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 774,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 91,
25
+ "prompt": "A realistic oil painting style depiction of Vincent van Gogh standing in his studio, mid-stroke as he paints on a large canvas. Van Gogh is wearing his iconic mustard yellow shirt and has a focused, intense expression. His palette and brushes are visible on an easel beside him. The room is filled with various paintings and sketches, giving it a cluttered yet artistic atmosphere. Soft sunlight filters through the window, casting a warm glow over the scene. The camera is positioned at a medium distance, capturing Van Gogh in a three-quarter profile view, emphasizing his dedication and passion for his work.",
26
+ "original_prompt": "Vincent van Gogh is painting in the room",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/091.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1217.6986980438232,
34
+ "excluded_context_kv_latency_ms": 788.0319442749023,
35
+ "wall_generation_s": 3.726020097732544,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.013439337722957134,
39
+ "psnr": 18.71622085571289,
40
+ "ssim": 0.7232256531715393,
41
+ "lpips": 0.15522535145282745,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/overall_consistency_092.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 775,
23
+ "prompt_suite": "overall_consistency",
24
+ "suite_index": 92,
25
+ "prompt": "A serene landscape featuring vibrant yellow flowers swaying gently in the breeze. The flowers are arranged in a field with patches of green grass, creating a soft and inviting backdrop. The sun is shining brightly, casting dappled shadows on the ground. A light wind causes the petals to flutter gracefully. The sky is clear and blue, with fluffy white clouds drifting by. The camera starts at a wide angle to capture the expansive field, then slowly zooms in to focus on the individual flowers as they dance in the wind. Medium close-up shot.",
26
+ "original_prompt": "Yellow flowers swing in the wind",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/overall_consistency/092.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1222.010729789734,
34
+ "excluded_context_kv_latency_ms": 790.2067184448242,
35
+ "wall_generation_s": 3.7348270416259766,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.06447853893041611,
39
+ "psnr": 11.905847787857056,
40
+ "ssim": 0.5747330188751221,
41
+ "lpips": 0.36539605259895325,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_000.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 776,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 0,
25
+ "prompt": "A dimly lit alleyway at night, filled with shadows and graffiti-covered walls. The ground is covered in wet cobblestones reflecting the faint glow of distant streetlights. In the background, there are tall buildings with their windows barely illuminated. A lone figure walks cautiously down the center of the alley, casting long shadows as they move. The atmosphere is tense and mysterious, with a slight mist hanging in the air. The camera follows the figure closely, maintaining a medium shot as they continue walking.",
26
+ "original_prompt": "alley",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/000.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1223.1961889266968,
34
+ "excluded_context_kv_latency_ms": 792.1530838012695,
35
+ "wall_generation_s": 3.7367825508117676,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.015905486419796944,
39
+ "psnr": 17.984530925750732,
40
+ "ssim": 0.5488008856773376,
41
+ "lpips": 0.3211682140827179,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_001.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 777,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 1,
25
+ "prompt": "A vibrant and bustling amusement park during the daytime, featuring colorful rides and attractions. The scene includes a large Ferris wheel, a roller coaster, bumper cars, and a carousel. People of various ages and ethnicities are enjoying themselves, smiling and laughing as they ride the attractions or walk around the park. The sun is shining brightly, casting warm light over the area. The background shows lush greenery and bright banners advertising different games and food stalls. Wide-angle shot capturing the entire lively atmosphere of the amusement park.",
26
+ "original_prompt": "amusement park",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/001.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1217.5069417953491,
34
+ "excluded_context_kv_latency_ms": 787.8211517333984,
35
+ "wall_generation_s": 3.7254958152770996,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.04260387644171715,
39
+ "psnr": 13.705508708953857,
40
+ "ssim": 0.4613979458808899,
41
+ "lpips": 0.4085935056209564,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_002.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 778,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 2,
25
+ "prompt": "A serene underwater scene in an aquarium, featuring a variety of colorful fish swimming gracefully among vibrant coral reefs and seaweed. Schools of small tropical fish dart around larger, more majestic creatures like angelfish and clownfish. In the background, soft sunlight filters through the water, creating gentle rays that illuminate the diverse marine life. The camera remains static, capturing the tranquil and peaceful atmosphere of the aquarium. Close-up underwater perspective.",
26
+ "original_prompt": "aquarium",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/002.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1218.5414819717407,
34
+ "excluded_context_kv_latency_ms": 790.311882019043,
35
+ "wall_generation_s": 3.73144268989563,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.05554395914077759,
39
+ "psnr": 12.553632259368896,
40
+ "ssim": 0.44553786516189575,
41
+ "lpips": 0.42000123858451843,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_003.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 779,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 3,
25
+ "prompt": "A grand architectural structure, showcasing intricate Gothic elements. The building features towering spires, detailed stone carvings, and large stained-glass windows depicting religious scenes. The exterior is bathed in warm evening sunlight, casting long shadows across the cobblestone street below. A gentle breeze causes the windows to sway slightly. The scene is captured from a mid-shot perspective, emphasizing the vertical lines and height of the structure.",
26
+ "original_prompt": "arch",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/003.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1228.6373453140259,
34
+ "excluded_context_kv_latency_ms": 795.2371444702148,
35
+ "wall_generation_s": 3.7460856437683105,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.04909453168511391,
39
+ "psnr": 13.08966875076294,
40
+ "ssim": 0.4721842110157013,
41
+ "lpips": 0.2912690043449402,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_004.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 780,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 4,
25
+ "prompt": "A beautifully lit art gallery featuring a variety of modern artworks displayed on pristine white walls. The gallery is moderately filled with visitors who are quietly admiring the pieces, some standing in front of paintings, others taking notes. Soft, ambient lighting casts a warm glow over the room, highlighting the textures and colors of the artwork. The floor is covered in sleek hardwood, and there are elegant wooden frames surrounding each piece. A few sculptures are also scattered throughout the space, adding depth and dimension to the gallery. Wide shots capturing the full expanse of the gallery, interspersed with close-ups focusing on individual artworks and the expressions of the visitors.",
26
+ "original_prompt": "art gallery",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/004.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1210.4287252426147,
34
+ "excluded_context_kv_latency_ms": 784.9614639282227,
35
+ "wall_generation_s": 3.7147607803344727,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03996914252638817,
39
+ "psnr": 13.982751369476318,
40
+ "ssim": 0.643336296081543,
41
+ "lpips": 0.31348899006843567,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_005.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 781,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 5,
25
+ "prompt": "A cozy, modern bathroom with warm lighting and sleek fixtures. The room features a large glass shower enclosure, a freestanding bathtub, and a double vanity with brushed nickel faucets. Soft, plush towels hang from a chrome towel rack. The walls are a calming light gray, and the floor is covered in elegant marble tiles. A frosted glass window lets in natural light, creating a bright and inviting atmosphere. Static wide shot showcasing the entire bathroom interior.",
26
+ "original_prompt": "bathroom",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/005.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1216.9472732543945,
34
+ "excluded_context_kv_latency_ms": 790.4086456298828,
35
+ "wall_generation_s": 3.7292473316192627,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.0035362979397177696,
39
+ "psnr": 24.51451063156128,
40
+ "ssim": 0.8576733469963074,
41
+ "lpips": 0.07928188890218735,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_006.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 782,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 6,
25
+ "prompt": "A cozy bakery shop interior with warm, inviting lighting. The space is filled with shelves displaying an array of freshly baked goods such as bread, pastries, and cakes. The counter is lined with cupcakes and cookies, while a glass case showcases artisanal bread loaves. A friendly baker in an apron is seen kneading dough at a large wooden table. Customers mill about, browsing and purchasing items. The walls are adorned with vintage posters and the scent of baking fills the air. Soft, ambient music plays in the background. Medium shot capturing the bustling atmosphere of the bakery.",
26
+ "original_prompt": "bakery shop",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/006.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1231.1879053115845,
34
+ "excluded_context_kv_latency_ms": 795.1878433227539,
35
+ "wall_generation_s": 3.7486820220947266,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.043915823101997375,
39
+ "psnr": 13.573789596557617,
40
+ "ssim": 0.47515714168548584,
41
+ "lpips": 0.3609626889228821,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_007.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 783,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 7,
25
+ "prompt": "A beautifully lit ballroom scene set during a grand ball. The room is filled with elegantly dressed guests, all engaged in graceful waltzing. The floor is polished hardwood, reflecting the chandeliers hanging from the ceiling. Ornate gold and white decorations adorn the walls, and lush curtains frame the large windows. Couples move smoothly across the dance floor, their dresses and tuxedos flowing as they twirl. The camera remains static, capturing the grandeur and elegance of the scene from a wide shot perspective.",
26
+ "original_prompt": "ballroom",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/007.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1215.132287979126,
34
+ "excluded_context_kv_latency_ms": 785.1604537963867,
35
+ "wall_generation_s": 3.7196974754333496,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.055082209408283234,
39
+ "psnr": 12.58988618850708,
40
+ "ssim": 0.46288251876831055,
41
+ "lpips": 0.3468609154224396,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_008.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 784,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 8,
25
+ "prompt": "A lively bar scene set during nighttime with warm, ambient lighting. The interior is decorated with dimly lit neon signs, wooden tables, and chairs, and a long polished bar counter with various bottles of alcohol neatly arranged behind it. Customers are engaged in conversations, some laughing, others quietly sipping their drinks. The bartender, a middle-aged man with a friendly smile, is attentively serving customers. The scene includes various races and ages of patrons, creating a diverse atmosphere. Medium shot capturing the bustling environment from a static perspective.",
26
+ "original_prompt": "bar",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/008.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1223.246042251587,
34
+ "excluded_context_kv_latency_ms": 790.8070678710938,
35
+ "wall_generation_s": 3.735969305038452,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.032396476715803146,
39
+ "psnr": 14.895021915435791,
40
+ "ssim": 0.47852325439453125,
41
+ "lpips": 0.3498237133026123,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_009.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 785,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 9,
25
+ "prompt": "A serene countryside barn at sunset, surrounded by lush green fields and tall golden wheat stalks swaying gently in the breeze. The barn has a rustic wooden exterior with a red roof and white trim, showcasing classic American farm architecture. The scene includes a few old farming tools leaning against the side of the barn. Soft sunlight filters through the clouds, casting a warm glow over the entire area. The background features rolling hills and distant trees. Static medium shot capturing the peaceful ambiance of the barn.",
26
+ "original_prompt": "barn",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/009.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1222.357458114624,
34
+ "excluded_context_kv_latency_ms": 792.7964172363281,
35
+ "wall_generation_s": 3.736220598220825,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.009277062490582466,
39
+ "psnr": 20.325894355773926,
40
+ "ssim": 0.6585938334465027,
41
+ "lpips": 0.16470643877983093,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_010.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 786,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 10,
25
+ "prompt": "A dimly lit basement interior with exposed brick walls and pipes running along the ceiling. The space is cluttered with old furniture, boxes, and various household items scattered around. A flickering fluorescent light casts eerie shadows across the room. The floor is covered with a mix of tiles and exposed concrete, adding to the gloomy atmosphere. The camera captures the entire room in a wide shot, focusing on the messiness and the general desolate feel of the basement.",
26
+ "original_prompt": "basement",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/010.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1211.2642097473145,
34
+ "excluded_context_kv_latency_ms": 786.6418304443359,
35
+ "wall_generation_s": 3.717756509780884,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.019418232142925262,
39
+ "psnr": 17.117903232574463,
40
+ "ssim": 0.5040470957756042,
41
+ "lpips": 0.32351383566856384,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_011.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 787,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 11,
25
+ "prompt": "A serene beach scene in vibrant, high-definition cinematography. The video captures the vast expanse of golden sand stretching towards the horizon, with gentle waves lapping at the shore. Palm trees sway gently in the breeze, their leaves rustling softly. Seagulls fly overhead, occasionally diving down to the water. The sun is setting, casting a warm, orange glow over the beach and reflecting off the calm sea. In the foreground, a lone figure walks along the shoreline, kicking up sand with each step. The background showcases a stunning array of pastel hues as the sky transitions from bright blue to shades of pink and purple. Wide shot, static camera capturing the panoramic beauty of the beach during sunset.",
26
+ "original_prompt": "beach",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/011.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1226.1798419952393,
34
+ "excluded_context_kv_latency_ms": 792.3141098022461,
35
+ "wall_generation_s": 3.741076946258545,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.02555770054459572,
39
+ "psnr": 15.924782752990723,
40
+ "ssim": 0.6802187561988831,
41
+ "lpips": 0.32937952876091003,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_012.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 788,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 12,
25
+ "prompt": "A cozy bedroom designed in a modern Scandinavian style. The room is softly lit with warm, ambient lighting from a bedside lamp. The walls are painted a calming light blue, and there is a plush, full-sized bed with a white duvet cover and soft pillows. A wooden nightstand holds the lamp and a small vase of fresh flowers. On the opposite wall, there is a large window with sheer curtains, allowing natural light to stream in during the day. The floor is covered with a soft, beige carpet. The scene is a medium shot, capturing the essence of a serene sleeping space.",
26
+ "original_prompt": "bedroom",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/012.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1226.32870388031,
34
+ "excluded_context_kv_latency_ms": 793.4942474365234,
35
+ "wall_generation_s": 3.7417430877685547,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.005390920210629702,
39
+ "psnr": 22.683370113372803,
40
+ "ssim": 0.8145282864570618,
41
+ "lpips": 0.08826501667499542,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_013.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 789,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 13,
25
+ "prompt": "A serene daytime view of a modern suspension bridge stretching across a wide river. The bridge is bustling with cars and pedestrians moving steadily along its pathways. The water below reflects the bridge's elegant curves and the surrounding green hills. Soft sunlight glints off the water, casting gentle ripples and highlighting the bridge's structural beauty. Wide shot, static scene focusing on the bridge's grandeur and the vibrant activity on and around it.",
26
+ "original_prompt": "bridge",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/013.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1214.3866682052612,
34
+ "excluded_context_kv_latency_ms": 787.0598068237305,
35
+ "wall_generation_s": 3.7218058109283447,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.014376369304955006,
39
+ "psnr": 18.42350721359253,
40
+ "ssim": 0.7122732400894165,
41
+ "lpips": 0.1753683239221573,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_014.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 790,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 14,
25
+ "prompt": "A serene botanical garden filled with a diverse array of vibrant flowers and lush greenery. The scene showcases various blooming flowers such as roses, lilies, and tulips, along with tall palm trees and dense shrubbery. Soft sunlight filters through the canopy, casting dappled shadows on the winding paths. A gentle breeze causes the leaves to rustle softly. In the background, there are quaint benches and water fountains adding to the tranquil atmosphere. The camera pans slowly across the garden, capturing the intricate details of each plant. Close-up shots of individual flowers highlight their colors and textures. Mid-shot of the garden path leading towards a beautiful fountain.",
26
+ "original_prompt": "botanical garden",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/014.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1226.9036922454834,
34
+ "excluded_context_kv_latency_ms": 792.3556900024414,
35
+ "wall_generation_s": 3.741184949874878,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03976272791624069,
39
+ "psnr": 14.005237817764282,
40
+ "ssim": 0.4430130422115326,
41
+ "lpips": 0.29480108618736267,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_015.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 791,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 15,
25
+ "prompt": "A bustling cafeteria scene in a modern school setting. Students and teachers are seen moving about, carrying trays and chatting. The cafeteria is brightly lit with large windows allowing natural sunlight to pour in. Various food stations are visible, including a salad bar, hot entrees, and a dessert area. The walls are painted in cheerful colors, adorned with posters and motivational quotes. In the background, a few students sit at round tables, enjoying their meals. The scene captures a lively atmosphere with a mix of diverse individuals, emphasizing natural interactions and movements. Wide shot, capturing the full environment.",
26
+ "original_prompt": "cafeteria",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/015.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1226.4226121902466,
34
+ "excluded_context_kv_latency_ms": 793.6547393798828,
35
+ "wall_generation_s": 3.7416718006134033,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.06467496603727341,
39
+ "psnr": 11.892638206481934,
40
+ "ssim": 0.5070303678512573,
41
+ "lpips": 0.3936202824115753,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_016.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 792,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 16,
25
+ "prompt": "A serene campsite nestled in a dense forest at sunset. The scene includes several tents pitched around a central campfire, where a group of diverse friends gather, laughing and sharing stories. The tents are brightly colored and varied in design, with camping gear scattered nearby. The sky is painted with warm hues of orange and pink, casting a gentle glow over the campsite. The ground is covered with soft green grass and fallen leaves. The atmosphere is cozy and inviting, capturing the essence of a perfect camping experience. Medium shot, static view.",
26
+ "original_prompt": "campsite",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/016.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1214.0344610214233,
34
+ "excluded_context_kv_latency_ms": 785.3417205810547,
35
+ "wall_generation_s": 3.718737840652466,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.02810785174369812,
39
+ "psnr": 15.511723756790161,
40
+ "ssim": 0.5571815371513367,
41
+ "lpips": 0.2314838320016861,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_017.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 793,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 17,
25
+ "prompt": "A vibrant college campus during a sunny afternoon, filled with students engaged in various activities. Young adults are walking between colorful buildings adorned with banners and posters, some are chatting while others are studying outdoors at picnic tables under large shade trees. The atmosphere is lively and energetic, with laughter and conversation filling the air. The scene includes a central quad with a fountain, surrounded by lush green lawns and blooming flowers. A few bicycles are parked near benches where students are relaxing. The background showcases modern architectural designs mixed with classic brick facades. Medium shot capturing the bustling environment of a typical university campus.",
26
+ "original_prompt": "campus",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/017.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1222.7482976913452,
34
+ "excluded_context_kv_latency_ms": 791.5199737548828,
35
+ "wall_generation_s": 3.736467123031616,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.051962848752737045,
39
+ "psnr": 12.843070030212402,
40
+ "ssim": 0.4117988348007202,
41
+ "lpips": 0.2886117696762085,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_018.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 794,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 18,
25
+ "prompt": "A vintage carousel at an amusement park, featuring brightly colored horses and other animals that move up and down as the carousel spins. The carousel is set against a backdrop of twinkling lights and festive decorations, creating a magical atmosphere. The horses have detailed designs, including intricate patterns and shimmering fabrics. Children and adults alike are seen riding the carousel, laughing and enjoying themselves. The scene is captured in a warm, nostalgic style, with soft lighting and a gentle camera pan following the rotation of the carousel. Medium shot focusing on the central action of the carousel in motion.",
26
+ "original_prompt": "carrousel",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/018.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1227.8457288742065,
34
+ "excluded_context_kv_latency_ms": 794.5906219482422,
35
+ "wall_generation_s": 3.744400978088379,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.048063553869724274,
39
+ "psnr": 13.181841373443604,
40
+ "ssim": 0.36238670349121094,
41
+ "lpips": 0.5201117396354675,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_019.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 795,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 19,
25
+ "prompt": "A majestic medieval castle standing tall against a dramatic sky filled with storm clouds. The castle is made of dark stone with towering turrets, intricate carvings, and banners fluttering from the walls. The moat surrounding the castle is calm and reflective, with a small bridge leading up to the entrance. The drawbridge is raised, and the castle gates are closed, emphasizing a sense of mystery and security. In the background, rolling hills covered in lush greenery can be seen. The scene is captured in a sweeping wide shot, capturing the grandeur and scale of the castle.",
26
+ "original_prompt": "castle",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/019.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1210.69722366333,
34
+ "excluded_context_kv_latency_ms": 783.6933212280273,
35
+ "wall_generation_s": 3.7137045860290527,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.015014189295470715,
39
+ "psnr": 18.234981298446655,
40
+ "ssim": 0.635593831539154,
41
+ "lpips": 0.18163959681987762,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_020.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 796,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 20,
25
+ "prompt": "A somber, misty cemetery at dusk, featuring tall tombstones covered in ivy and moss. The graves are evenly spaced, surrounded by weathered iron fences and overgrown grass. A single path winds through the center, leading the viewer deeper into the graveyard. The sky is a deep shade of orange and purple, with a few stars beginning to twinkle. In the distance, a lone tree casts long shadows across the ground. The atmosphere is quiet and reflective, with a soft, eerie glow from the setting sun casting long shadows. Static wide shot capturing the entire scene.",
26
+ "original_prompt": "cemetery",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/020.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1224.9022388458252,
34
+ "excluded_context_kv_latency_ms": 791.2953567504883,
35
+ "wall_generation_s": 3.738220691680908,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.009804723784327507,
39
+ "psnr": 20.085647106170654,
40
+ "ssim": 0.6244985461235046,
41
+ "lpips": 0.27084049582481384,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_021.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 797,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 21,
25
+ "prompt": "A realistic classroom scene set during a typical school day. The classroom has rows of desks facing a chalkboard at the front. Students are engaged in various activities; some are reading books, others are writing in notebooks, and a few are quietly talking. The teacher stands at the front of the class, holding a book and addressing the students. The room is well-lit with sunlight streaming in from large windows, casting soft shadows across the desks. The walls are adorned with educational posters and motivational quotes. Medium shot capturing the full classroom environment.",
26
+ "original_prompt": "classroom",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/021.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1231.0826416015625,
34
+ "excluded_context_kv_latency_ms": 795.9347457885742,
35
+ "wall_generation_s": 3.7498552799224854,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03873363137245178,
39
+ "psnr": 14.119117259979248,
40
+ "ssim": 0.5727826356887817,
41
+ "lpips": 0.2854697108268738,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_022.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 798,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 22,
25
+ "prompt": "A dramatic, panoramic view of a rugged cliff face overlooking the ocean. The cliff is covered in layers of weathered rock, with deep crevices and moss-covered boulders scattered along its surface. Jagged cliffs rise up from the water, creating a dramatic silhouette against the sky. Waves crash against the base of the cliff, sending sprays of water into the air. The sky is filled with dramatic clouds casting shadows across the rocky terrain. Wide shot, static scene focusing on the vastness and power of nature.",
26
+ "original_prompt": "cliff",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/022.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1212.54167842865,
34
+ "excluded_context_kv_latency_ms": 785.3398742675781,
35
+ "wall_generation_s": 3.7174222469329834,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03478219732642174,
39
+ "psnr": 14.586429595947266,
40
+ "ssim": 0.5196868181228638,
41
+ "lpips": 0.3075571656227112,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_023.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 799,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 23,
25
+ "prompt": "A bustling city street intersection at rush hour, featuring a well-worn crosswalk painted in bright white stripes. Pedestrians of various ages and ethnicities move across the crosswalk, some in a hurry, others leisurely walking their pets or talking on phones. Cars and buses stop at the curb, waiting patiently for the pedestrians to clear before proceeding. The background showcases tall buildings with neon signs flickering in the evening light. The scene is captured from a mid-shot perspective, focusing on the dynamic interactions between pedestrians and vehicles.",
26
+ "original_prompt": "crosswalk",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/023.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1219.6274137496948,
34
+ "excluded_context_kv_latency_ms": 791.0679473876953,
35
+ "wall_generation_s": 3.732987403869629,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.0480739064514637,
39
+ "psnr": 13.18090558052063,
40
+ "ssim": 0.45218610763549805,
41
+ "lpips": 0.3372775912284851,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_024.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 800,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 24,
25
+ "prompt": "A bustling construction site during daytime, featuring multiple cranes lifting heavy materials, workers in bright orange safety vests and hard hats operating machinery and tools. The site includes tall steel beams, concrete mixers, and piles of bricks and gravel. Dust rises from various activities, giving the scene a gritty, industrial feel. The background shows partially constructed buildings and a skyline of modern skyscrapers. Wide shot capturing the entire lively construction area, with a focus on the dynamic motion of workers and machines.",
26
+ "original_prompt": "construction site",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/024.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1221.9253425598145,
34
+ "excluded_context_kv_latency_ms": 792.3005447387695,
35
+ "wall_generation_s": 3.7362589836120605,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.031271710991859436,
39
+ "psnr": 15.048483610153198,
40
+ "ssim": 0.5574068427085876,
41
+ "lpips": 0.30211982131004333,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_025.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 801,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 25,
25
+ "prompt": "A long, dimly lit corridor in a gothic-style building, with high ceilings and arched windows letting in soft, diffused sunlight. The walls are lined with ornate wooden panels and old oil paintings in gold frames. The floor is covered in worn-out marble tiles, and there are flickering sconces providing additional lighting along the sides. Shadows dance across the walls as the camera slowly pans down the corridor, revealing various doorways leading to unknown rooms. The atmosphere is eerie yet intriguing, inviting viewers to explore further. Static shot, medium distance.",
26
+ "original_prompt": "corridor",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/025.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1212.1812629699707,
34
+ "excluded_context_kv_latency_ms": 785.1160278320312,
35
+ "wall_generation_s": 3.7178640365600586,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03444795310497284,
39
+ "psnr": 14.628366231918335,
40
+ "ssim": 0.47727787494659424,
41
+ "lpips": 0.31977832317352295,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_026.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 802,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 26,
25
+ "prompt": "A serene courtyard during early morning, featuring a well-manicured lawn with dew-covered grass. The courtyard includes a central fountain with water gently flowing, surrounded by neatly arranged flower beds filled with vibrant blooms. Stone pathways wind through the area, leading to a small seating area with wooden benches. Tall trees provide shade, their leaves rustling softly in the breeze. The sky is a soft pastel hue, with sunlight filtering through the branches. The scene is calm and peaceful, captured in a medium shot that showcases the entire courtyard.",
26
+ "original_prompt": "courtyard",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/026.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1219.223747253418,
34
+ "excluded_context_kv_latency_ms": 790.4981079101562,
35
+ "wall_generation_s": 3.731290578842163,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.020918913185596466,
39
+ "psnr": 16.79460883140564,
40
+ "ssim": 0.5845323204994202,
41
+ "lpips": 0.19777199625968933,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_027.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 803,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 27,
25
+ "prompt": "A vast, arid desert landscape under a scorching sun. Rolling sand dunes stretch as far as the eye can see, their peaks catching the sunlight and casting deep shadows. Sparse patches of dry vegetation dot the sandy terrain. In the distance, rugged rocky formations rise from the sea of sand. The sky is a clear, intense blue with wisps of clouds. A gentle breeze stirs the fine grains of sand, creating subtle waves across the dune slopes. Wide shot capturing the expansive and desolate beauty of the desert.",
26
+ "original_prompt": "desert",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/027.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1222.6551351547241,
34
+ "excluded_context_kv_latency_ms": 792.5663604736328,
35
+ "wall_generation_s": 3.737196922302246,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.018288973718881607,
39
+ "psnr": 17.378106117248535,
40
+ "ssim": 0.661460816860199,
41
+ "lpips": 0.2115335613489151,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_028.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 804,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 28,
25
+ "prompt": "A bustling downtown area during the daytime, featuring tall skyscrapers with reflective glass facades, pedestrians walking along busy sidewalks, and vehicles moving through crowded streets. Include details such as street vendors, advertising billboards, and people in various postures and expressions. The scene captures the vibrant energy and diversity of urban life. Use a realistic documentary style with medium shots to capture the interactions between people and their environment. Static scene with no camera movement.",
26
+ "original_prompt": "downtown",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/028.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1211.1251707077026,
34
+ "excluded_context_kv_latency_ms": 785.3196792602539,
35
+ "wall_generation_s": 3.7163097858428955,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.026118066161870956,
39
+ "psnr": 15.830589532852173,
40
+ "ssim": 0.4947797954082489,
41
+ "lpips": 0.26805466413497925,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_029.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 805,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 29,
25
+ "prompt": "A serene morning scene of a suburban driveway, lined with neatly trimmed bushes and flower beds. A cobblestone path leads up to a two-story house with a white picket fence in the background. The driveway is empty except for a single parked car, casting a shadow under the bright sun. The camera starts with a wide shot, then slowly zooms in to focus on the textured surface of the driveway, highlighting the subtle play of light and shadow. The background features a clear blue sky and lush greenery. Medium-long shot.",
26
+ "original_prompt": "driveway",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/029.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1223.227168083191,
34
+ "excluded_context_kv_latency_ms": 790.8925476074219,
35
+ "wall_generation_s": 3.736015558242798,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.035228852182626724,
39
+ "psnr": 14.531015157699585,
40
+ "ssim": 0.45182862877845764,
41
+ "lpips": 0.318172812461853,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_030.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 806,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 30,
25
+ "prompt": "A serene rural farm scene during the golden hour. The farm includes a large red barn with white trim, a few scattered hay bales, and lush green fields dotted with grazing cows. In the foreground, a farmer in overalls and a straw hat is working diligently, tending to crops. The sky is painted with soft hues of orange and pink, casting a warm glow over the landscape. The background features rolling hills and distant forests. Mid-shot capturing the farmer and the immediate farm environment.",
26
+ "original_prompt": "farm",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/030.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1225.7045392990112,
34
+ "excluded_context_kv_latency_ms": 794.6262664794922,
35
+ "wall_generation_s": 3.7423653602600098,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.007835183292627335,
39
+ "psnr": 21.059508323669434,
40
+ "ssim": 0.6760666966438293,
41
+ "lpips": 0.15915295481681824,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_031.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 807,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 31,
25
+ "prompt": "A bustling food court in a modern shopping mall during the lunch rush. Multiple food stalls are visible, each with unique signage and colorful displays of various cuisines such as pizza, sushi, burgers, and noodles. Customers are walking around, browsing, and queuing up at different counters. Some sit at round tables scattered throughout the area, enjoying their meals. The environment is vibrant with bright lighting, tiled floors, and glass partitions separating the seating from the food stalls. Medium shots capturing the energy and variety of the food court from multiple angles.",
26
+ "original_prompt": "food court",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/031.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1212.870213508606,
34
+ "excluded_context_kv_latency_ms": 785.4048233032227,
35
+ "wall_generation_s": 3.7186279296875,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.060686029493808746,
39
+ "psnr": 12.169113159179688,
40
+ "ssim": 0.43265312910079956,
41
+ "lpips": 0.3916473984718323,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_032.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 808,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 32,
25
+ "prompt": "A panoramic view of a well-maintained football field under clear blue skies. The grass is lush and green, with white sidelines marking the boundaries. A goal post stands at each end, and the midfield line runs across, dividing the field into two equal halves. Players in vibrant jerseys are scattered across the field, engaged in various natural motions such as running, passing the ball, and defending. Spectators fill the stands, cheering loudly. The camera captures the expansive field from a high angle, showcasing the entirety of the playing area. Wide shot, static scene.",
26
+ "original_prompt": "football field",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/032.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1216.3387813568115,
34
+ "excluded_context_kv_latency_ms": 787.6956558227539,
35
+ "wall_generation_s": 3.724975824356079,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.017414048314094543,
39
+ "psnr": 17.591001987457275,
40
+ "ssim": 0.6593030095100403,
41
+ "lpips": 0.18224118649959564,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_033.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 809,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 33,
25
+ "prompt": "A serene forest road winding through a dense woodland, covered in soft moss and fallen leaves. The path is flanked by towering trees with gnarled trunks and lush green canopies above. Sunlight filters through the leaves, casting dappled shadows on the ground. A gentle mist rises from the forest floor, adding to the mystical atmosphere. In the background, the road stretches into the distance, inviting viewers to explore further. The scene is captured in a medium shot, with a static camera to emphasize the tranquil beauty of nature.",
26
+ "original_prompt": "forest road",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/033.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1221.665831565857,
34
+ "excluded_context_kv_latency_ms": 791.957878112793,
35
+ "wall_generation_s": 3.7350785732269287,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.03550906106829643,
39
+ "psnr": 14.496607780456543,
40
+ "ssim": 0.4292260706424713,
41
+ "lpips": 0.30203568935394287,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_034.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 810,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 34,
25
+ "prompt": "A serene daytime scene of a classic European fountain in a bustling city square, surrounded by cobblestone pathways and lined with blooming flowers. The fountain features intricate stone carvings and a central spout from which water cascades gracefully into a basin below. The water sparkles in the sunlight, creating a tranquil atmosphere. People walk past the fountain, some pausing to admire the water flow. The background includes tall buildings and passersby, adding to the lively urban environment. Medium shot, static scene.",
26
+ "original_prompt": "fountain",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/034.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1212.0753860473633,
34
+ "excluded_context_kv_latency_ms": 785.613899230957,
35
+ "wall_generation_s": 3.7166988849639893,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.06547371298074722,
39
+ "psnr": 11.839330196380615,
40
+ "ssim": 0.40923991799354553,
41
+ "lpips": 0.3060734272003174,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_035.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 811,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 35,
25
+ "prompt": "A realistic daytime scene at a gas station, featuring a typical American-style convenience store with bright neon signs and pumps. The store has large glass windows and a green roof, surrounded by paved parking lots with cars parked nearby. Inside the store, various products are displayed on shelves, including snacks, drinks, and cigarettes. A middle-aged man in a blue work uniform stands at the counter, smiling warmly as he interacts with a customer. Outside, a woman fills up her car with gasoline at one of the fuel pumps. The background shows a few trees and a clear blue sky. Wide shot, static camera.",
26
+ "original_prompt": "gas station",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/035.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1221.4552297592163,
34
+ "excluded_context_kv_latency_ms": 790.9777526855469,
35
+ "wall_generation_s": 3.7345452308654785,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.00557349668815732,
39
+ "psnr": 22.538721561431885,
40
+ "ssim": 0.8565146327018738,
41
+ "lpips": 0.08093481510877609,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_036.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 812,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 36,
25
+ "prompt": "Aerial time-lapse footage of a massive glacier in a stunning mountain range. The glacier is covered in layers of snow and ice, with deep blue crevices and towering ice walls. Ice chunks break off and fall into a calm, crystal-clear glacial lake below, creating small ripples that spread across the surface. The surrounding area is filled with lush greenery and tall, majestic peaks. The sky transitions from bright daylight to the soft hues of twilight, casting a serene glow over the landscape. Wide shot, static scene capturing the vastness and beauty of the natural environment.",
26
+ "original_prompt": "glacier",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/036.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1227.1288738250732,
34
+ "excluded_context_kv_latency_ms": 795.0322875976562,
35
+ "wall_generation_s": 3.7447574138641357,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.01951812393963337,
39
+ "psnr": 17.095619440078735,
40
+ "ssim": 0.6219541430473328,
41
+ "lpips": 0.1694403439760208,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_037.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 813,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 37,
25
+ "prompt": "A serene midday scene on a lush golf course with rolling green hills and pristine fairways. The grass is neatly trimmed, and there are tall palm trees scattered throughout the landscape. In the foreground, a group of three golfers, two men and a woman, are standing beside a golf ball tee-up area, each holding a golf club. They wear casual golf attire, including polo shirts and khaki pants. The man on the left is addressing the ball, preparing to swing, while the other two observe him intently. The background showcases a clear blue sky with fluffy clouds, and the distant horizon reveals more of the expansive golf course. The scene is captured in a wide-angle shot, emphasizing the vastness and beauty of the course.",
26
+ "original_prompt": "golf course",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/037.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1210.5083904266357,
34
+ "excluded_context_kv_latency_ms": 788.3092193603516,
35
+ "wall_generation_s": 3.718170642852783,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.007943176664412022,
39
+ "psnr": 21.00005865097046,
40
+ "ssim": 0.7696701288223267,
41
+ "lpips": 0.11208619177341461,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }
eval_out/per_prompt/cfa_atc_chunk_s2_fppp/scene_038.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "complete",
3
+ "protocol": "Self-Forcing Extended-251 Full Evaluation",
4
+ "strategy": "cfa_atc_chunk_s2_fppp",
5
+ "base_model": "causal_forcing_a",
6
+ "method": "predictor_atc_chunk_s2",
7
+ "param": "pattern",
8
+ "param_value": "FPPP",
9
+ "target_speedup": 2.6415094339622645,
10
+ "schedule": "FPPP",
11
+ "predictor_config": {
12
+ "atc_freq_dim": 256,
13
+ "atc_gate_hidden_dim": 512,
14
+ "atc_gate_initial_probability": 0.3,
15
+ "atc_mlp_hidden_dim": 3072,
16
+ "atc_previous_scope": "chunk",
17
+ "atc_transport_residual_scale": 0.1,
18
+ "input_variant": "atc",
19
+ "source_layer": 17,
20
+ "weights": "/local/zoubin/cz/projects/Causal-Forcing-a/output/layer17_stage2_dmd_atc_chunk_lora128_4gpu_b2_2000steps/checkpoint_step_2000/predictor_ema.safetensors"
21
+ },
22
+ "global_index": 814,
23
+ "prompt_suite": "scene",
24
+ "suite_index": 38,
25
+ "prompt": "An indoor gymnasium filled with various exercise equipment and athletes in mid-action. The space is bright with ample lighting, featuring parallel bars, weightlifting stations, and a running track. Athletes are seen performing natural motions such as lifting weights, running, and stretching. The background includes spectators sitting on bleachers, cheering for the athletes. The gymnasium has a modern design with large windows and sleek architecture. Wide shot capturing the entire gymnasium from an elevated angle to showcase the layout and activities.",
26
+ "original_prompt": "indoor gymnasium",
27
+ "seed": 0,
28
+ "video": "/local/zoubin/cz/projects/comparison/eval_out/generated_videos/cfa_atc_chunk_s2_fppp/scene/038.mp4",
29
+ "num_frames": 81,
30
+ "height": 480,
31
+ "width": 832,
32
+ "fps": 16,
33
+ "policy_latency_ms": 1224.4151802062988,
34
+ "excluded_context_kv_latency_ms": 789.8335571289062,
35
+ "wall_generation_s": 3.735917806625366,
36
+ "reference_strategy": "cfa_ffff",
37
+ "pixel_metrics_vs_ffff": {
38
+ "mean_mse": 0.030477792024612427,
39
+ "psnr": 15.160164833068848,
40
+ "ssim": 0.5376057624816895,
41
+ "lpips": 0.2877686023712158,
42
+ "num_frames": 81
43
+ },
44
+ "cache_diagnostics": {
45
+ "denoise_forwards": 28,
46
+ "full_forwards": 10,
47
+ "predictor_forwards": 18,
48
+ "predictor_compute_equivalent": 0.03333333333333333,
49
+ "compute_equivalent_forwards": 10.6,
50
+ "middle_steps": 14,
51
+ "middle_compute_equivalent": -3.4000000000000004
52
+ }
53
+ }