bundle:
  repository: Orangerl/umm
  visibility: public
  stage: stage2_covt
  depth_token_count: 4
  created_utc: '2026-08-06'

data:
  rows: 889489
  images: 756894
  source_json: /rczhang/rczhang/zhangrch/video_gen/gen_dataset/covt/data.json
  source_images: /rczhang/rczhang/zhangrch/video_gen/gen_dataset/covt/images
  note: Local CoVT composition; image references are rewritten to portable basenames.

checkpoints:
  stage1_alignment:
    source: /rczhang/rczhang/zhangrch/video_gen/gen_checkpoint/stage1_algin_v2/embeddings_connector/checkpoint-32000.bin
  qwen_base:
    source: /rczhang/rczhang/zhangrch/video_gen/pretrined_models/Qwen/Qwen3-VL-4B-Instruct
  sam:
    source: /rczhang/rczhang/zhangrch/video_gen/pretrined_models/covt_moe/sam_vit_h_4b8939.pth
  vggt:
    source: facebook/VGGT-1B
    revision: 860abec7937da0a4c03c41d3c269c366e82abdf9
    upstream_source_commit: a288dd0f14786c93483e45524328726ab7b1b4ce
  dinov2:
    source: facebookresearch/dinov2
    entrypoint: dinov2_vitl14
    processor_source: facebook/dinov2-large
    processor_revision: 47b73eefe95e8d44ec3623f8890bd894b6ea2d6c
  pidinet:
    source_file: table5_baseline.pth
  siglip:
    source: google/siglip2-large-patch16-256
    revision: 787800c8990e6f058423089178e718139608408c

editable_sources:
  transformers:
    commit: 374d44d54adb1c5f52e68aff97d1675f56d657a8
    local_modifications_included: true
  diffusers:
    commit: 065e36937a5f69a0018a7238d07aaf087345df4c
    local_modifications_included: false

environment:
  source_prefix: /rczhang/rczhang/conda_packs/unvideo
  archive: environment/unvideo-conda.tar.gz
  format: conda-pack tar.gz
  python: '3.10'
  pytorch: 2.8.0+cu128
  deepspeed: 0.18.9