本文へ移動
cccskills
無料GitHub で公開

fengshenbang-lm

Operate the Fengshenbang-LM Fengshen package for pipelines, model families, data/training utilities, examples, and conversion planning.

VectorSpaceLab/AREX-Skill
インストール方法を見る

含まれるファイル（42）

  • SKILL.md3.8 KB
  • references/installation.md3.0 KB
  • references/repo-provenance.md1.8 KB
  • references/repo-routing-metadata.json548 B
  • references/troubleshooting.md3.6 KB
  • scripts/check_fengshen_install.py2.4 KB
  • sub-skills/data-training/references/data-formats.md7.2 KB
  • sub-skills/data-training/references/distributed-training.md5.1 KB
  • sub-skills/data-training/references/metrics-and-validation.md4.0 KB
  • sub-skills/data-training/references/pretraining-workflows.md6.1 KB
  • sub-skills/data-training/references/training-arguments.md7.5 KB
  • sub-skills/data-training/references/troubleshooting.md5.4 KB
  • sub-skills/data-training/scripts/check_ner_labels.py4.8 KB
  • sub-skills/data-training/scripts/inspect_training_args.py7.0 KB
  • sub-skills/data-training/SKILL.md2.7 KB
  • sub-skills/examples-conversion/references/clue-and-task-recipes.md6.8 KB
  • sub-skills/examples-conversion/references/conversion-utilities.md8.6 KB
  • sub-skills/examples-conversion/references/model-family-recipes.md7.7 KB
  • sub-skills/examples-conversion/references/nlg-nlt-recipes.md9.0 KB
  • sub-skills/examples-conversion/references/taiyi-diffusion.md7.9 KB
  • sub-skills/examples-conversion/references/troubleshooting.md8.7 KB
  • sub-skills/examples-conversion/references/ziya-llama.md9.3 KB
  • sub-skills/examples-conversion/scripts/check_recipe_requirements.py3.7 KB
  • sub-skills/examples-conversion/scripts/plan_ziya_conversion.py3.0 KB
  • sub-skills/examples-conversion/SKILL.md5.7 KB
  • sub-skills/model-zoo/references/api-reference.md10.0 KB
  • sub-skills/model-zoo/references/auto-and-tokenizers.md9.0 KB
  • sub-skills/model-zoo/references/compatibility.md7.4 KB
  • sub-skills/model-zoo/references/model-overview.md9.5 KB
  • sub-skills/model-zoo/references/troubleshooting.md10.4 KB
  • sub-skills/model-zoo/scripts/check_model_imports.py11.0 KB
  • sub-skills/model-zoo/SKILL.md3.5 KB
  • sub-skills/pipelines-cli/references/cli-reference.md5.4 KB
  • sub-skills/pipelines-cli/references/sequence-tagging.md4.5 KB
  • sub-skills/pipelines-cli/references/tcbert.md3.6 KB
  • sub-skills/pipelines-cli/references/text-classification.md5.2 KB
  • sub-skills/pipelines-cli/references/troubleshooting.md6.0 KB
  • sub-skills/pipelines-cli/references/unimc-uniex-ubert.md5.5 KB
  • sub-skills/pipelines-cli/scripts/inspect_pipeline_cli.py8.9 KB
  • sub-skills/pipelines-cli/scripts/make_classification_fixture.py5.5 KB
  • sub-skills/pipelines-cli/scripts/make_sequence_tagging_fixture.py4.6 KB
  • sub-skills/pipelines-cli/SKILL.md3.2 KB

SKILL.md（原文）

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Fengshenbang-LM repo skill

Use this skill when the task involves the Fengshenbang-LM / fengshen Python package: Chinese NLP model families, fengshen-pipeline, Fengshen pipeline APIs, dataloaders, PyTorch Lightning training helpers, Deepspeed/Megatron planning, model-family examples, Taiyi/Ziya recipes, or checkpoint conversion planning.

First checks

  1. Read references/installation.md before installing or repairing the package. This repo is sensitive to the age of the Transformers, Lightning, Torchmetrics, NumPy, and Setuptools stack.

  2. Run the safe install check when an environment is available:

    python scripts/check_fengshen_install.py --check-cli-help

    This imports package modules and inspects CLI help only; it does not download models or launch training.

  3. If imports, CLI help, optional CUDA, model downloads, submodules, or hard-coded example paths fail, read references/troubleshooting.md.

  4. Use references/repo-provenance.md to decide whether this skill matches the current package checkout or should be refreshed.

Route by task

User taskRead
Use fengshen-pipeline, inspect pipeline CLI flags, create tiny classification/NER fixtures, adapt text classification or sequence-tagging data, use UniMC/UniEX/Ubert/TCBert APIssub-skills/pipelines-cli/SKILL.md
Choose or import Fengshen model/config/tokenizer families, understand top-level exports, custom auto/config keys, model compatibility, or model import failuressub-skills/model-zoo/SKILL.md
Prepare/validate data, inspect UniversalDataModule, metrics, checkpoints, optimizer/scheduler/Trainer flags, pretraining data, Deepspeed/Megatron/fused-kernel planningsub-skills/data-training/SKILL.md
Plan CLUE/classification recipes, NLG/NLT examples, Taiyi Stable Diffusion, Ziya inference/fine-tuning/conversion, delta/TF/diffusers/LLaMA conversion utilitiessub-skills/examples-conversion/SKILL.md

Package facts to remember

  • Distribution/import package: fengshen.
  • Version in package metadata: 0.0.1.
  • Console entry point: fengshen-pipeline = fengshen.cli.fengshen_pipeline:main.
  • CLI pattern: fengshen-pipeline <pipeline_module_name> predict|train ...; source CLI supports only predict and train methods.
  • Public top-level exports include Longformer, RoFormer, Megatron-T5, and Ubert classes/pipelines.
  • Many examples are real but heavy: they may require model/data downloads, CUDA/Deepspeed, multi-GPU launchers, large memory, checkpoint mutation, or maintainer-local paths. Treat them as planning evidence unless the user explicitly authorizes backend verification and execution.

Non-goals and safety

  • Do not run original repo examples, pretraining scripts, stable-diffusion scripts, Ziya conversion scripts, or fused-kernel tests by default.
  • Do not tell the user to open files in the original repository checkout as runtime instructions. Use this skill's bundled references and scripts.
  • Do not claim CUDA/Deepspeed/large-model execution is verified from CPU import checks.
  • Ask before network downloads, large model/dataset cache use, GPU allocation, multi-node jobs, checkpoint conversion, or overwriting output directories.
  • If a task is about editing or refreshing this generated skill after the source repo changes, use the repo-skill refresh workflow rather than relying on stale guidance.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

ログインしてレビューを書く

同じリポジトリのスキル

概要と使いどころ

3d-resnets-pytorch

無料

Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.

日本語の概要は準備中です。原文の説明を表示しています。

VectorSpaceLab/AREX-Skill3322026年9月3日 更新

3ddfa

無料

Guide 3DDFA Python inference, geometry rendering, training/evaluation, and optional C++ ONNX workflows for 3D dense face alignment.

日本語の概要は準備中です。原文の説明を表示しています。

VectorSpaceLab/AREX-Skill3322026年9月3日 更新

3ddfa-v2

無料

Routes 3DDFA_V2 face-alignment setup, still-image demos, video tracking, and ONNX benchmarking workflows.

日本語の概要は準備中です。原文の説明を表示しています。

VectorSpaceLab/AREX-Skill3322026年9月3日 更新

ab3dmot

無料

Operate AB3DMOT 3D multi-object tracking workflows for KITTI and nuScenes data, tracking, evaluation, and visualization.

日本語の概要は準備中です。原文の説明を表示しています。

VectorSpaceLab/AREX-Skill3322026年9月3日 更新

accelerate

無料

Use Hugging Face Accelerate for PyTorch training-loop migration, distributed launch/configuration, DeepSpeed/FSDP/TPU backend setup, big-model inference/offload, checkpointing, tracking, and troubleshooting.

日本語の概要は準備中です。原文の説明を表示しています。

VectorSpaceLab/AREX-Skill3322026年9月3日 更新

acme

無料

Route Acme reinforcement-learning framework tasks across core loops, replay/data, JAX agents, and TensorFlow/Sonnet agents.

日本語の概要は準備中です。原文の説明を表示しています。

VectorSpaceLab/AREX-Skill3322026年9月3日 更新

VectorSpaceLab のスキルをすべて見る

このスキルの問題を報告する