StreamDiffusionV2/run_v2v.sh at master · DigiField/StreamDiffusionV2 · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
#!/bin/bash
set -eu

# Unified offline launcher for all supported video-to-video inference modes.
# Environment variables can override the default config, checkpoint, output, GPU,
# and launch parameters without having to edit the script itself.
MODE="${1:-single}"
if [ "$#" -gt 0 ]; then
  shift
fi

ROOT_DIR="$(CDPATH= cd -- "$(dirname "$0")" && pwd)"
MODEL_TYPE="${MODEL_TYPE:-T2V-1.3B}"
PREV_ARG=""
for ARG in "$@"; do
  if [ "$PREV_ARG" = "--model_type" ]; then
    MODEL_TYPE="$ARG"
    break
  fi
  case "$ARG" in
    --model_type=*)
      MODEL_TYPE="${ARG#--model_type=}"
      break
      ;;
  esac
  PREV_ARG="$ARG"
done

if [ "$MODEL_TYPE" = "T2V-14B" ]; then
  CONFIG_PATH="${CONFIG_PATH:-configs/wan_causal_dmd_v2v_14b.yaml}"
  CHECKPOINT_FOLDER="${CHECKPOINT_FOLDER:-ckpts/wan_causal_dmd_v2v_14b}"
  STEP="${STEP:-1}"
else
  CONFIG_PATH="${CONFIG_PATH:-configs/wan_causal_dmd_v2v.yaml}"
  CHECKPOINT_FOLDER="${CHECKPOINT_FOLDER:-ckpts/wan_causal_dmd_v2v}"
  STEP="${STEP:-2}"
fi
OUTPUT_FOLDER="${OUTPUT_FOLDER:-outputs/}"
HEIGHT="${HEIGHT:-480}"
WIDTH="${WIDTH:-832}"
FPS="${FPS:-16}"
MASTER_PORT="${MASTER_PORT:-29501}"

case "$MODE" in
  single|batch)
    # Standard single-GPU batched inference.
    CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0}" python -m streamv2v.inference \
      --config_path "$ROOT_DIR/$CONFIG_PATH" \
      --checkpoint_folder "$ROOT_DIR/$CHECKPOINT_FOLDER" \
      --output_folder "$ROOT_DIR/$OUTPUT_FOLDER" \
      --prompt_file_path "$ROOT_DIR/${PROMPT_FILE_PATH:-examples/prompt.txt}" \
      --video_path "$ROOT_DIR/${VIDEO_PATH:-examples/original.mp4}" \
      --height "$HEIGHT" \
      --width "$WIDTH" \
      --fps "$FPS" \
      --step "$STEP" \
      --model_type "$MODEL_TYPE" \
      "$@"
    ;;
  single-wo|wo|wo-batch)
    # Single-GPU inference without batched denoising.
    CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0}" python -m streamv2v.inference_wo_batch \
      --config_path "$ROOT_DIR/$CONFIG_PATH" \
      --checkpoint_folder "$ROOT_DIR/$CHECKPOINT_FOLDER" \
      --output_folder "$ROOT_DIR/$OUTPUT_FOLDER" \
      --prompt_file_path "$ROOT_DIR/${PROMPT_FILE_PATH:-examples/prompt.txt}" \
      --video_path "$ROOT_DIR/${VIDEO_PATH:-examples/original.mp4}" \
      --height "$HEIGHT" \
      --width "$WIDTH" \
      --fps "$FPS" \
      --step "$STEP" \
      --model_type "$MODEL_TYPE" \
      "$@"
    ;;
  pipe|parallel)
    # Pipeline-parallel inference across multiple GPUs on one node.
    CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0,1}" torchrun \
      --nproc_per_node="${NPROC_PER_NODE:-2}" \
      --master_port="$MASTER_PORT" \
      --module streamv2v.inference_pipe \
      --config_path "$ROOT_DIR/$CONFIG_PATH" \
      --checkpoint_folder "$ROOT_DIR/$CHECKPOINT_FOLDER" \
      --output_folder "$ROOT_DIR/$OUTPUT_FOLDER" \
      --prompt_file_path "$ROOT_DIR/${PROMPT_FILE_PATH:-examples/prompt.txt}" \
      --video_path "$ROOT_DIR/${VIDEO_PATH:-examples/original.mp4}" \
      --height "$HEIGHT" \
      --width "$WIDTH" \
      --fps "$FPS" \
      --step "$STEP" \
      --model_type "$MODEL_TYPE" \
      "$@"
    ;;
  *)
    echo "Unknown mode '$MODE'. Expected one of: single, single-wo, pipe" >&2
    exit 1
    ;;
esac