Skip to content

Labels

Labels

46 labels

  • API

    API
    related to API changes

  • app: colossal-eval

    app: colossal-eval
    applications/ColossalEval: LLM evaluation

  • app: colossal-llama

    app: colossal-llama
    applications/Colossal-LLaMA: continual pretraining on LLaMA

  • app: colossalchat

    app: colossalchat
    applications/ColossalChat (formerly Coati): SFT, RLHF, DPO, GRPO

  • bug

    bug
    Something isn't working

  • cannotreplicate

    cannotreplicate
    Maintainers cannot reproduce it from the description

  • code reviewing

    code reviewing

  • community-request

    community-request
    Requested by an external user

  • compatibility

    compatibility
    related to compatibility

  • diffusion

    diffusion
    stable diffusion related

  • documentation

    documentation
    Improvements or additions to documentation

  • duplicate

    duplicate
    This issue or pull request already exists

  • enhancement

    enhancement
    New feature or request

  • good first issue

    good first issue
    Good for newcomers

  • help wanted

    help wanted
    Extra attention is needed

  • in-progress

    in-progress
    Someone is working on it or a PR is open; exempt from stale

  • invalid

    invalid
    This doesn't seem right

  • known issue

    known issue
    known issue to be fixed

  • module: auto-parallel

    module: auto-parallel
    Automatic parallel strategy search (largely unmaintained since 2023)

  • module: booster/plugin

    module: booster/plugin
    Booster and its plugins: TorchDDP, LowLevelZero, Gemini, HybridParallel, TorchFSDP

  • module: checkpoint

    module: checkpoint
    Saving and loading model / optimizer states, sharded checkpoints, Hugging Face format

  • module: ci

    module: ci
    CI, GitHub Actions, test infrastructure and the tests themselves

  • module: examples

    module: examples
    Examples, tutorials and benchmark scripts under examples/

  • module: fp8/quantization

    module: fp8/quantization
    FP8 training and communication, low-precision quantization

  • module: inference

    module: inference
    Inference engine (colossalai/inference)

  • module: install/build

    module: install/build
    Installation, extension builds, wheels and environment problems

  • module: kernel

    module: kernel
    CUDA / Triton kernels and extensions: build, correctness, performance

  • module: lazy-init

    module: lazy-init
    LazyInit: build an empty model and materialize parameters after sharding

  • module: moe

    module: moe
    MoE models and expert parallelism: expert sharding, routing, Mixtral/DeepSeek/Qwen MoE

  • module: optimizer

    module: optimizer
    Optimizers and LR schedulers: CPUAdam, HybridAdam, FusedAdam, LAMB, GaLore