MindAct

candle-orgApache-2.0活跃维护

让 HuggingFace 上的 Transformers 和 Diffusers 模型无缝跑在昇腾 MindSpore 上,省去手动迁移适配的麻烦,并支持加速。

MindSpore + 🤗Huggingface: Run any Transformers/Diffusers model on MindSpore with seamless compatibility and acceleration.

v0.5.1发布于 2025-11-05 07:34 UTC920star269fork同步于 2026-11-07 07:31 UTC

怎么上手

由 README 提炼,执行前请确认
pip install -e .
mindact config-check configs/experiments/libero-baseline.yaml
硬件要求按 README 说明

README 未说明

下载:源码 zip·GitHub Releases(安装包)本站不镜像安装包

项目说明

由 README 自动翻译提炼

面向桌面操作任务的模仿学习策略训练与评估框架,整合 LeRobot 数据集与 LIBERO 仿真基准

让 HuggingFace 上的 Transformers 和 Diffusers 模型无缝跑在昇腾 MindSpore 上,省去手动迁移适配的麻烦,并支持加速。

核心能力

  • 用适配器接口统一数据集、策略与环境
  • YAML 配置加 manifest 记录实验来源,可复现
  • torch、lerobot、libero 按 extras 懒加载
  • 约定目录结构管理 checkpoint、日志与结果
  • 内置 fake runner 跑通评估流程做冒烟测试

容易踩的坑

v0.1 只是骨架版本,真实 LeRobot 策略与 LIBERO 环境需等适配器实现,fake runner 不是基准结果;torch、lerobot、libero 都是可选依赖,需按需安装对应 extras;仓库由 MindNLP 转型而来,旧代码在 legacy 分支,别照旧文档用

内容由模型根据 README 生成,可能不完整或不准确,以官方文档为准。看原文 README

最近发版

5 个版本v0.5.1
  1. v0.5.1最新正式版2025-11-05 · 2025-11-05 07:34 UTC

    这个版本没有写更新说明。

  2. v0.5.02025-11-03 · 2025-11-03 07:42 UTC
    查看更新说明

    What's Changed

  3. v0.4.12025-04-10 · 2025-04-10 07:55 UTC
    查看更新说明

    What's Changed

  4. v0.4.02024-10-14 · 2024-10-14 03:54 UTC
    查看更新说明

    时隔5个月,MindNLP 0.4来啦!!!这一次在开源社区同学们的努力下带来了数个重大特性更新!!!

    1. 模型大量更新,支持250+ transformers模型,覆盖90% huggingface模型,使用MindSpore进行模型复现、开发训练再无迁移之忧(社区会同步huggingface和业界新模型,随时更新)。
    2. 全平台支持:全面支持昇腾910A、910B、310B(香橙派)、GPU、CPU。(PS:目前香橙派唯一可用的AI开发套件)
    3. 分布式并行推理:支持多卡多进程并行推理(超过10B的模型可以玩啦!)
    4. 量化算法支持:香橙派支持Smooth Quant,GPU支持bitsandbytes int8量化
    5. Sentence transformer支持:可以开心的搞RAG啦
    6. 动态图性能优化:昇腾硬件下动态图推理速度达到Pytorch+GPU同水平(Llama实测性能85ms/token)
    7. 真正的动静统一:通过mindspore.jit一行切换图模式,完全使用huggingface code style,既保证易用性,又能快速提升性能。昇腾硬件下Llama实测性能为动态图2倍(45ms/token),与mindspore其他静态图based套件性能完全一致。
    8. 海量LLM应用更新:包括文本信息抽取、聊天机器人、语音识别、ChatPDF、音乐生成、代码生成、声音克隆等等,伴随着模型支持度的提高,更多好玩的应用等你来开发!

    欢迎大家使用、提Issue和Star一下~

    安装方法:

    pip install mindnlp
    

    代码仓: https://github.com/mindspore-lab/mindnlp

    PS:获取昇腾算力的几个途径:

  5. v0.3.12024-05-17 · 2024-05-17 02:08 UTC
    查看更新说明

    What's Changed

    New Contributors

    Full Changelog: https://github.com/mindspore-lab/mindnlp/compare/v0.3.0...v0.3.1

用户点评

登录用户的真实使用记录,可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译

    MindAct

    A PyTorch and Hugging Face native training and evaluation toolkit for embodied policies

    MindAct is a reproducible training and evaluation framework for imitation learning policies on desktop manipulation tasks. It integrates LeRobot datasets and policies with LIBERO simulation benchmarks, providing experiment provenance tracking and standardized evaluation protocols.

    Features

    • Protocol-based architecture: Minimal adapter interfaces for datasets, policies, and environments
    • Reproducible experiments: YAML configurations with manifest-tracked provenance
    • Optional dependencies: Lazy loading of torch, lerobot, and libero via extras
    • Artifact management: Conventional directory structure for checkpoints, logs, and results
    • Type-safe: Frozen dataclasses and runtime protocols throughout

    Quick Start

    Installation

    # Basic installation (configuration and CLI only)
    pip install -e .
    
    # With PyTorch
    pip install -e ".[torch]"
    
    # With LeRobot datasets and policies
    pip install -e ".[lerobot]"
    
    # With LIBERO simulation environments
    pip install -e ".[libero]"
    
    # All integrations
    pip install -e ".[torch,lerobot,libero]"
    
    # Development tools
    pip install -e ".[dev]"
    

    Run an Example

    # Validate configuration
    mindact config-check configs/experiments/libero-baseline.yaml
    
    # Run the dependency-free evaluation smoke path
    mindact eval configs/experiments/libero-baseline.yaml \
      --runner fake \
      --run-id smoke-run \
      --output-dir outputs \
      --episodes 2
    

    config-check validates a configuration, and eval --runner fake exercises the full evaluation lifecycle with built-in test doubles. Real LeRobot policies and LIBERO environments arrive with their adapter implementations; the fake runner is a contract smoke test, not a benchmark result.

    Configuration Format

    name: libero-baseline
    seed: 42
    output_dir: outputs
    
    dataset:
      repo_id: lerobot/aloha_sim_insertion_human
      revision: null
      split: train
    
    policy:
      name: act
      pretrained_model: null
    
    environment:
      name: libero
      task_suite: libero_spatial
      task_ids: []
    
    training:
      steps: 10000
      batch_size: 8
      learning_rate: 0.0001
      log_every: 100
      checkpoint_every: 2000
    
    evaluation:
      episodes: 20
      max_steps: 500
      record_video: false
    

    Project Status

    MindAct v0.1 is the initial skeleton release. Core interfaces and configuration system are stable. Implementation priorities:

    Phase B (current): Unified reproducible training and evaluation
    Phase A (future): Trajectory quality diagnostics

    See docs/ for architecture details and contribution guidelines.

    Repository History

    This repository was originally MindNLP, a MindSpore-based NLP library. The legacy codebase is preserved in the legacy branch. MindAct represents a complete pivot to embodied AI with PyTorch and Hugging Face as the native stack.

    License

    Apache License 2.0. See LICENSE and NOTICE for details.