API Reference¶
The Bit-Axon API is organized into the following subpackages.
Public API¶
The top-level bit_axon package exports the core model classes:
Subpackages¶
| Subpackage | Description |
|---|---|
Config | Model configuration dataclass |
Model | 24-layer sandwich model |
Tokenizer | HuggingFace tokenizer wrapper |
Training | Training pipeline, LoRA, DoRA, ORPO |
Inference | Autoregressive generation |
Layers | Core neural network layers |
Quantization | NF4 weight quantization |
Profiling | Memory, speed, thermal profiling |
Evaluation | Perplexity computation |
See also¶
- Architecture Overview — Model design and configuration details
- CLI Reference — Command-line interface for all operations
- Training Guide — End-to-end fine-tuning walkthrough
- Inference Guide — Generation, chat, and streaming