Video · video-tokenization · class-conditional-image-generation · autoregressive-generation

Open-MAGVIT2

TencentARC

Runtime source is vendored in-tree and the CUDA generation wrapper is wired.

Integrated统一环境open-magvit2

构建运行命令

选择仓库中已记录的 variant，然后复制对应的准备、检查或运行命令。

worldfoundry.pipeline

Variant

任务video-tokenization

环境worldfoundry-unified-cu128

设备CUDA 11.3

worldfoundry-eval evaluate \
  --mode model \
  --model-id open-magvit2 \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/open-magvit2 \
  --metric artifact_count \
  --json

兼容性与版本

由 Manifest 生成

集成状态: integrated
Runner 证据: checkpoint backed verified
环境: worldfoundry-unified-cu128
Python: 3.10
CUDA: CUDA 11.3
PyTorch: torch
源码 revision: —
Checkpoint revision: 607230591be9efcb4ec5ec9b6417b9e863f2dd2c

Runtime profile: open-magvit2
Pipeline binding: open-magvit2
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.component_pipelines:OpenMAGVIT2Pipeline
Backend stage: in_tree_runtime
Runtime 状态: checkpoint_backed_verified
Driver 状态: compatible_with_downgraded_torch
环境类型: 统一环境

安装环境

环境解析器会读取已记录 profile，并选择此处显示的统一或独立环境。

bash scripts/setup/model_env_install.sh --model open-magvit2

依赖版本约束 15

torch
torchvision
lightning==2.6.1
jsonargparse[signatures]
numpy
tensorboard
tensorboardx
albumentations
omegaconf
einops
requests
transformers
lpips
av
decord

Conda 依赖 3

python
pip
ffmpeg

Checkpoint 与资产

分配算力前先做本地检查；gated、private 与 license 字段直接来自 checkpoint manifest。

worldfoundry-eval zoo model-download --model-id open-magvit2 --check-local --json

TencentARC/Open-MAGVIT2

Revision: 607230591be9efcb4ec5ec9b6417b9e863f2dd2c
License: apache-2.0
Gated: —
Private: —

TencentARC/Open-MAGVIT2

Revision: —
License: —
Gated: —
Private: —

运行与输出

生成的命令通过共享 evaluation 边界运行，并持久保存结果 manifest 与 artifact。

worldfoundry-eval evaluate \
  --mode model \
  --model-id open-magvit2 \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/open-magvit2 \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`class_id`	Required
`image`	Optional
`prompt`	Optional

Artifact 契约

Artifact 类型	文件名 / 路径
`generated_image`	`open_magvit2.png`

证据与来源

Catalog 集成、原生 demo parity 与 runner parity 是三条独立记录。

集成状态: integrated
Runner 证据: checkpoint backed verified
原生 Demo 证据: 未记录
验证 Imports: torch, lightning, transformers, decord

配方溯源

worldfoundry/data/models/catalog/video/open-magvit2.yamlGitHubhttps://github.com/TencentARC/Open-MAGVIT2

构建运行命令

输入契约

Artifact 契约

相关配方