| 1 | # ============================================================================= |
| 2 | # Echo-LongVideo Multishot Inference Configuration |
| 3 | # |
| 4 | # Used by: python inference.py --config configs/inference.yaml |
| 5 | # Command-line arguments override values in this file. |
| 6 | # ============================================================================= |
| 7 | |
| 8 | # --------------------------------------------------------------------------- |
| 9 | # Environment |
| 10 | # --------------------------------------------------------------------------- |
| 11 | env: |
| 12 | venv_path: .venv |
| 13 | |
| 14 | # --------------------------------------------------------------------------- |
| 15 | # Paths (relative paths are resolved from repository root) |
| 16 | # --------------------------------------------------------------------------- |
| 17 | paths: |
| 18 | checkpoint: checkpoints/test.safetensors |
| 19 | gemma_path: checkpoints/gemma-3-12b |
| 20 | prompts_dir: prompts |
| 21 | prompts_glob: "*.json" |
| 22 | output_root: inference_result/outputs |
| 23 | |
| 24 | # --------------------------------------------------------------------------- |
| 25 | # Video / generation settings |
| 26 | # --------------------------------------------------------------------------- |
| 27 | video: |
| 28 | num_frames: 241 |
| 29 | height: 736 |
| 30 | width: 1280 |
| 31 | fps: 25 |
| 32 | seed: 12345 |
| 33 | |
| 34 | # --------------------------------------------------------------------------- |
| 35 | # Denoising schedule |
| 36 | # --------------------------------------------------------------------------- |
| 37 | denoising: |
| 38 | steps: |
| 39 | - 1000 |
| 40 | - 994 |
| 41 | - 988 |
| 42 | - 981 |
| 43 | - 975 |
| 44 | - 909 |
| 45 | - 725 |
| 46 | - 422 |
| 47 | - 0 |
| 48 | sigmas: |
| 49 | - 1.0 |
| 50 | - 0.99375 |
| 51 | - 0.9875 |
| 52 | - 0.98125 |
| 53 | - 0.975 |
| 54 | - 0.909375 |
| 55 | - 0.725 |
| 56 | - 0.421875 |
| 57 | - 0.0 |
| 58 | |
| 59 | # --------------------------------------------------------------------------- |
| 60 | # Memory bank (paired audio-video memory) |
| 61 | # --------------------------------------------------------------------------- |
| 62 | memory: |
| 63 | max_size: 7 |
| 64 | num_fix_frames: 3 |
| 65 | downscale_factor: 1 |
| 66 | position_mode: reference |
| 67 | lora_strength: 1.0 |
| 68 | lora_generator: true |
| 69 | lora_path: "" |
| 70 | save_mode: random_every_shot_frame |
| 71 | frame_selection_mode: center |
| 72 | clip_num_frames: 9 |
| 73 | |
| 74 | # --------------------------------------------------------------------------- |
| 75 | # Audio memory |
| 76 | # --------------------------------------------------------------------------- |
| 77 | audio_memory: |
| 78 | enable: true |
| 79 | window_size: 96 |
| 80 | window_selection_mode: max_response |
| 81 | sample_rate: 16000 |
| 82 | mel_bins: 128 |
| 83 | mel_hop_length: 160 |
| 84 | n_fft: 1024 |
| 85 | downsample_factor: 4 |
| 86 | is_causal: true |
| 87 | |
| 88 | # --------------------------------------------------------------------------- |
| 89 | # Inference runtime |
| 90 | # --------------------------------------------------------------------------- |
| 91 | inference: |
| 92 | device: cuda |
| 93 | dtype: bfloat16 |
| 94 | v2a_grad_scale: 2.0 |
| 95 |