Embodied · vla · vla.policy_rollout · vlm_policy · robot_policy

RoboFlamingo

robovlms

RoboFlamingo/OpenFlamingo architecture and inference source are vendored in-tree; public pipeline CUDA action_trace validation passed.

Runner verified统一环境roboflamingo

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

环境worldfoundry-unified-cu128

设备Prepare Only

worldfoundry-eval evaluate \
  --mode model \
  --model-id roboflamingo \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/roboflamingo \
  --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: 3.8
CUDA: Prepare Only
PyTorch: torch
源码 revision: —
Checkpoint revision: —

Runtime profile: roboflamingo
Pipeline binding: roboflamingo
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.component_pipelines:RoboFlamingoPipeline
Backend stage: in_tree_runtime
Runtime 状态: in_tree_roboflamingo_predict_action_gpu_ready
Driver 状态: prepare_only_vlm_policy_runtime_requires_checkpoint_validation
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 roboflamingo

依赖版本约束 14

numpy
Pillow
opencv-python
torch
torchvision
torchaudio
transformers
huggingface-hub
open-clip-torch
einops
einops-exts
pytorch-lightning
omegaconf
safetensors

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 roboflamingo --check-local --json

robovlms/RoboFlamingo

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

openflamingo/OpenFlamingo-3B-vitl-mpt1b-langinstruct

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

anas-awadalla/mpt-1b-redpajama-200b-dolly

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

输入契约

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

Artifact 契约

Artifact 类型	文件名 / 路径
`action_trace`	`—`
`action_trace`	`roboflamingo_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, cv2, torch, transformers, open_clip

配方溯源

worldfoundry/data/models/catalog/vla_va_wam/roboflamingo.yamlGitHubhttps://github.com/RoboFlamingo/RoboFlamingo

构建运行命令

输入契约

Artifact 契约

相关配方