Embodied · vla · vla.action_prediction · vla.policy_rollout

OpenVLA

openvla

In-tree OpenVLA HF runtime is registered through the WorldFoundry pipeline binding.

Runner verified统一环境源码已固定openvla

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vla.action_prediction

环境worldfoundry-unified-cu128

设备CUDA 11.3

worldfoundry-eval evaluate \
  --mode model \
  --model-id openvla-7b \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/openvla-7b \
  --metric artifact_count \
  --json

兼容性与版本

由 Manifest 生成

集成状态: integrated
Runner 证据: verified
环境: worldfoundry-unified-cu128
Python: 3.10
CUDA: CUDA 11.3
PyTorch: torch
源码 revision: 47a0ec7fc4ec123775a391911046cf33cf9ed83f
Checkpoint revision: 47a0ec7fc4ec123775a391911046cf33cf9ed83f

Runtime profile: openvla
Pipeline binding: openvla
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.component_pipelines:OpenVLAPipeline
Backend stage: in_tree_runtime
Runtime 状态: in_tree_openvla_predict_action_ready
Driver 状态: compatible
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 openvla-7b

依赖版本约束 10

numpy
Pillow
transformers
accelerate
safetensors
sentencepiece
torch
torchvision
timm
tokenizers

Conda 依赖 2

python
pip

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 openvla --check-local --json

openvla/openvla-7b

Revision: 47a0ec7fc4ec123775a391911046cf33cf9ed83f
License: mit
Gated: —
Private: —

openvla/openvla-7b-finetuned-libero-10

Revision: 80970322773f81baa2e22fe495d0487b93a05cfa
License: mit
Gated: —
Private: —

openvla/openvla-7b

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

openvla/openvla-7b-finetuned-libero-10

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 openvla-7b \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/openvla-7b \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`actions`	robot_action, proprio, action_space
`image`	Required
`instruction`	Required
`prompt`	Required
`video`	Optional

Artifact 契约

Artifact 类型	文件名 / 路径
`actions`	`—`
`action_trace`	`—`
`rollout_metrics`	`—`
`action_trace`	`openvla_action_trace.json`

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: verified
原生 Demo 证据: 未记录
验证 Imports: numpy, PIL, torch, transformers

配方溯源

worldfoundry/data/models/catalog/vla_va_wam/openvla.yamlGitHubhttps://github.com/openvla/openvlaHugging Face47a0ec7fc4ec123775a391911046cf33cf9ed83fHugging Face80970322773f81baa2e22fe495d0487b93a05cfa

构建运行命令

输入契约

Artifact 契约

相关配方