Upload lingbot-world-720p/report.md with huggingface_hub
Browse files- lingbot-world-720p/report.md +52 -0
lingbot-world-720p/report.md
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# LingBot-World Inference Reproduction Report
|
| 2 |
+
|
| 3 |
+
## Setup
|
| 4 |
+
|
| 5 |
+
- Repository: https://github.com/robbyant/lingbot-world
|
| 6 |
+
- Conda environment: `lingbot_world_repro`
|
| 7 |
+
- FlashAttention: `2.7.4.post1` prebuilt wheel for CUDA 12 / PyTorch 2.5 / Python 3.10.
|
| 8 |
+
- GPU execution: 1 Slurm node, `torchrun --nproc_per_node=4`
|
| 9 |
+
- Generation size argument: `720*1280` with source aspect ratio preserved.
|
| 10 |
+
- Frame count: 81 frames at 16 FPS
|
| 11 |
+
- Normal inference sample steps override: default
|
| 12 |
+
- Control visualization: outputs generated with `action_path` include WASD/IJKL-style control UI; `fast_inference` uses the same visualization path added to `WanI2VFast`.
|
| 13 |
+
|
| 14 |
+
## Method Difference
|
| 15 |
+
|
| 16 |
+
- `inference` runs `generate.py` with the LingBot-World Base camera model. It denoises the full video sequence with the normal Wan I2V sampling path.
|
| 17 |
+
- `fast_inference` runs `generate_fast.py` with the LingBot-World Fast weights under `lingbot_world_fast`. It uses causal chunk-by-chunk inference and KV caching to reduce latency for interactive generation.
|
| 18 |
+
|
| 19 |
+
## Aggregate Metrics
|
| 20 |
+
|
| 21 |
+
| Method | Completed Videos | Mean Time | Peak Single GPU VRAM | Peak Total VRAM |
|
| 22 |
+
| --- | ---: | ---: | ---: | ---: |
|
| 23 |
+
| inference | 8 | 43m 45.0s | 61.9 GiB | 242.7 GiB |
|
| 24 |
+
| fast_inference | 8 | 8m 52.9s | 59.9 GiB | 178.5 GiB |
|
| 25 |
+
|
| 26 |
+
## Comparison
|
| 27 |
+
|
| 28 |
+
- Mean wall time: `fast_inference` is 4.93x faster than `inference` on this 8-case run.
|
| 29 |
+
- Peak single-GPU VRAM: `fast_inference` reduces the observed peak by 2.0 GiB.
|
| 30 |
+
- Peak total VRAM across 4 GPUs: `fast_inference` reduces the observed peak by 64.2 GiB.
|
| 31 |
+
- Video validation: see the generated MP4 files and metrics CSV for output paths and sizes.
|
| 32 |
+
|
| 33 |
+
## Per-Video Results
|
| 34 |
+
|
| 35 |
+
| Method | Case | Status | Time | Peak Single GPU VRAM | Output |
|
| 36 |
+
| --- | --- | --- | ---: | ---: | --- |
|
| 37 |
+
| fast_inference | case00_seed142 | ok | 12m 48.0s | 59.9 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case00_seed142.mp4` |
|
| 38 |
+
| fast_inference | case00_seed42 | ok | 8m 47.3s | 58.5 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case00_seed42.mp4` |
|
| 39 |
+
| fast_inference | case01_seed143 | ok | 3m 22.0s | 58.7 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case01_seed143.mp4` |
|
| 40 |
+
| fast_inference | case01_seed43 | ok | 3m 20.1s | 58.7 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case01_seed43.mp4` |
|
| 41 |
+
| fast_inference | case02_seed44 | ok | 13m 23.2s | 59.0 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case02_seed44.mp4` |
|
| 42 |
+
| fast_inference | case03_seed45 | ok | 9m 41.7s | 57.4 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case03_seed45.mp4` |
|
| 43 |
+
| fast_inference | case04_seed46 | ok | 6m 56.7s | 57.2 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case04_seed46.mp4` |
|
| 44 |
+
| fast_inference | case05_seed47 | ok | 12m 44.1s | 58.7 GiB | `experiments/lingbot_720p/outputs/fast_inference/fast_inference_case05_seed47.mp4` |
|
| 45 |
+
| inference | case00_seed142 | ok | 46m 05.4s | 60.5 GiB | `experiments/lingbot_720p/outputs/inference/inference_case00_seed142.mp4` |
|
| 46 |
+
| inference | case00_seed42 | ok | 41m 18.5s | 60.5 GiB | `experiments/lingbot_720p/outputs/inference/inference_case00_seed42.mp4` |
|
| 47 |
+
| inference | case01_seed143 | ok | 47m 16.7s | 61.9 GiB | `experiments/lingbot_720p/outputs/inference/inference_case01_seed143.mp4` |
|
| 48 |
+
| inference | case01_seed43 | ok | 44m 22.1s | 60.3 GiB | `experiments/lingbot_720p/outputs/inference/inference_case01_seed43.mp4` |
|
| 49 |
+
| inference | case02_seed44 | ok | 45m 31.5s | 60.3 GiB | `experiments/lingbot_720p/outputs/inference/inference_case02_seed44.mp4` |
|
| 50 |
+
| inference | case03_seed45 | ok | 46m 22.2s | 59.7 GiB | `experiments/lingbot_720p/outputs/inference/inference_case03_seed45.mp4` |
|
| 51 |
+
| inference | case04_seed46 | ok | 38m 22.4s | 60.2 GiB | `experiments/lingbot_720p/outputs/inference/inference_case04_seed46.mp4` |
|
| 52 |
+
| inference | case05_seed47 | ok | 40m 41.4s | 60.5 GiB | `experiments/lingbot_720p/outputs/inference/inference_case05_seed47.mp4` |
|