Embodied · vla · vla.action_prediction · vla.policy_rollout · robot_policy · subtask_prediction

GigaBrain-0

open-gigaai

WorldFoundry in-tree GigaBrain-0.1 runtime emitted a checkpoint-backed CUDA action_trace using converted LeRobot RobotWin stats.

Runner verified统一环境giga-brain-0

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 12.8

worldfoundry-eval evaluate \
  --mode model \
  --model-id giga-brain-0-3.5b-base \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/giga-brain-0-3.5b-base \
  --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: 3.11
CUDA: CUDA 12.8
PyTorch: torch
源码 revision: —
Checkpoint revision: —

Runtime profile: giga-brain-0
Pipeline binding: giga-brain-0
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.component_pipelines:GigaBrain0Pipeline
Backend stage: in_tree_runtime
Runtime 状态: in_tree_giga_brain_0_runtime_converted_lerobot_stats_predict_gpu_ready_non_leaderboard
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 giga-brain-0-3.5b-base

依赖版本约束 12

numpy
Pillow
torch
torchvision
matplotlib
PyYAML
numpydantic
tyro
transformers
accelerate
safetensors
opencv-python

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 giga-brain-0 --check-local --json

open-gigaai/GigaBrain-0-3.5B-Base

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

open-gigaai/GigaBrain-0.1-3.5B-Base

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

google/paligemma-3b-pt-224

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

physical-intelligence/fast

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 giga-brain-0-3.5b-base \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/giga-brain-0-3.5b-base \
  --metric artifact_count \
  --json

输入契约

字段	已记录契约
`action_chunk`	50
`actions`	robot_action, action_chunk, continuous_action, subtask, trajectory_2d
`data_path`	lerobot_dataset
`embodiment_id`	Required
`image`	Required
`instruction`	Required
`original_action_dim`	robot_specific
`prompt`	Required
`state`	Required
`video`	Optional

Artifact 契约

Artifact 类型	文件名 / 路径
`action_trace`	`—`
`action_trace`	`giga_brain_0_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, torchvision, tyro, matplotlib, yaml, transformers, accelerate, safetensors, cv2

配方溯源

worldfoundry/data/models/catalog/vla_va_wam/giga-brain-0.yamlGitHubhttps://github.com/open-gigaai/giga-brain-0Hugging Facehttps://huggingface.co/open-gigaai/GigaBrain-0-3.5B-BaseHugging Facehttps://huggingface.co/open-gigaai/GigaBrain-0.1-3.5B-BaseHugging Facehttps://huggingface.co/google/paligemma-3b-pt-224Hugging Facehttps://huggingface.co/physical-intelligence/fast

构建运行命令

输入契约

Artifact 契约

相关配方