Package uno.anahata.asi.huggingface

package uno.anahata.asi.huggingface
Provides the provider implementation for the Hugging Face Inference API.

This module extends the OpenAI-compatible stack but adds a "Deep Inspection" layer. On model discovery, it queries the Hugging Face Hub directly to fetch:

  • Architecture: To detect reasoning fields (e.g. DeepSeek's reasoning_content).
  • Tokenizers: To detect R1-style tags (e.g. <think>) in templates.
  • Configuration: To derive accurate context windows and output limits.

This ensures that even generic inference endpoints gain full ASI capability (reasoning, tool calling) by understanding the underlying model's DNA.

Author:
anahata
  • Classes
    Class
    Description
    HuggingFaceModel
    A specialized model instance that encapsulates its own metadata discovery from the Hugging Face Hub.
    HuggingFaceProvider
    A specialized provider for the Hugging Face Inference API that performs deep model inspection by fetching metadata directly from the HF Hub.