Kenhuang Academy
  • English

      • Resources

      Transformers and Hub APIs move quickly—prefer the official docs. This page lists primary documentation, related courses on this site, and an optional path for coding agents.

      Official

      ResourceURL
      Transformers docshttps://huggingface.co/docs/transformers
      Transformers GitHubhttps://github.com/huggingface/transformers
      Hub modelshttps://huggingface.co/models
      Hub datasetshttps://huggingface.co/datasets
      Spaceshttps://huggingface.co/spaces
      NLP Coursehttps://huggingface.co/learn/nlp-course
      PEFT / LoRAhttps://huggingface.co/docs/peft
      datasetshttps://huggingface.co/docs/datasets
      Acceleratehttps://huggingface.co/docs/accelerate
      huggingface_hub / hf CLIhttps://huggingface.co/docs/huggingface_hub
      Hugging Face Skillshttps://huggingface.co/docs/hub/en/agents-skills

      Skills let Cursor, Claude Code, and similar tools follow SKILL.md for Hub / training jobs. That is not the main path of this tutorial.

      Docs you will reopen

      TopicWhere (concept)
      First tourQuicktour / Pipeline
      Load tipsfrom_pretrained, device_map="auto", dtype="auto"
      Generationgenerate(), GenerationConfig, max_new_tokens
      TrainingTrainer, TrainingArguments
      Three objectsPreTrainedConfig, PreTrainedModel, Preprocessor
      CourseLinkWhen to open it
      PyTorch/en/pytorch/Tensors, nn.Module, handwritten training loops
      Ollama/en/ollama/Easier local chat without owning generate()
      LangChain/en/langchain/Tools, agents, RAG (example 2 is intentionally retrieval-free)

      A useful order: PyTorch for foundations → Transformers to load and fine-tune Hub models → Ollama for everyday chat.

      Tutorial index

      1. Hugging Face tutorial
      2. Introduction
      3. Installation
      4. Quick Start
      5. Pipeline
      6. Tokenizers and Models
      7. Text Generation
      8. Datasets
      9. Fine-tuning
      10. The Hub
      11. Practical Examples

      Suggested exercises

      1. Classify 10 reviews you wrote; note the score distribution
      2. Compare DistilGPT2 greedy vs temperature=0.9 continuations
      3. Run Trainer.train() on a 500-row rotten_tomatoes slice
      4. (GPU) Attach LoRA to Qwen 0.5B and print_trainable_parameters()
      5. push_to_hub a local folder and add one paragraph to the model card

      Contribute

      This tutorial lives in Kenhuang Academy. When APIs change, open an issue or PR and update both docs/zh/huggingface/ and docs/en/huggingface/.

      For loading, follow the official tips: prefer device_map="auto" and dtype="auto".

      评论