Video · video-generation · memory · stream · interactive-world-model

VMem

liguang0115

In-tree VMem runtime at worldfoundry/synthesis/visual_generation/vmem/vmem_runtime; no external checkout is required.

Runner verified统一环境vmem

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-generation

环境worldfoundry-unified-cu128

设备CUDA 12.8

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

兼容性与版本

由 Manifest 生成

集成状态: verified
Runner 证据: verified
环境: worldfoundry-unified-cu128
Python: 3.11
CUDA: CUDA 12.8
PyTorch: torch
源码 revision: —
Checkpoint revision: ac5921080a57f5a634f4b9acbbc8f3db67c9d113

Runtime profile: vmem
Pipeline binding: vmem
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.vmem.pipeline_vmem:VMemPipeline
Backend stage: official_in_tree_runtime
Runtime 状态: verified
Driver 状态: compatible_unified_default
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 vmem

依赖版本约束 4

torch
torchvision
torchaudio
xfuser

Conda 依赖 5

python=3.11
pip
ffmpeg
setuptools
wheel

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 vmem --check-local --json

liguang0115/vmem

Revision: ac5921080a57f5a634f4b9acbbc8f3db67c9d113
License: —
Gated: —
Private: —

liguang0115/cut3r

Revision: b14faf986da0df405cff1b41e60e2975c4da2745
License: —
Gated: —
Private: —

SagiPolaczek/stable-diffusion-2-1-base

Revision: d1dc907291ce017defe5ec44dc1ec7ef55a1e26e
License: —
Gated: —
Private: —

laion/CLIP-ViT-H-14-laion2B-s32B-b79K

Revision: 1c2b8495b28150b8a4922ee1c8edee224c284c0c
License: —
Gated: —
Private: —

naver/DUSt3R_ViTLarge_BaseDecoder_512_dpt

Revision: 61c57447d7b0adc8a1a30b2b0adec7a8935aa2a3
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 vmem \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/vmem \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`prompt`	Required
`image`	Required
`video`	Optional
`actions`	interaction, navigation

Artifact 契约

Artifact 类型	文件名 / 路径
`generated_video`	`vmem.mp4`

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: verified
Runner 证据: verified
原生 Demo 证据: 未记录
验证 Imports: torch, torchvision, numpy, transformers, diffusers, xfuser, worldfoundry

配方溯源

worldfoundry/data/models/catalog/video/vmem.yamlProjecthttps://v-mem.github.io/Paperhttps://arxiv.org/abs/2506.18903GitHubhttps://github.com/runjiali-rl/vmemHugging Facehttps://huggingface.co/liguang0115/vmemHugging Facehttps://huggingface.co/liguang0115/cut3r

构建运行命令

输入契约

Artifact 契约

相关配方