---
license: mit
language:
  - en
tags:
  - text-to-speech
  - chatterbox
  - lora
  - onnx
base_model: ResembleAI/chatterbox
---

# Chatterbox-Turbo with LoRA input ports

Re-exported Chatterbox-Turbo ONNX with 192 optional LoRA adapter input
ports added to the language_model graph. Enables per-speaker fine-tuning
via ORT's `Ort::LoraAdapter` API.

## Files

- `language_model_q4f16.onnx` — graph file (~277 KB)
- `language_model_q4f16.onnx_data` — external weights file (~700 MB,
  fp32 — quantization to q4f16 is a TODO)

## Modifications vs upstream

- 96 additive LoRA branches (`MatMul → MatMul → Add`) spliced into each
  GPT2 block's c_attn / c_proj / c_fc / mlp.c_proj projections
- 192 named optional graph inputs: `lora.layers.{N}.{parent}_{child}.{A|B}`
- Zero-sized default initializers so the un-adapted graph runs identically
  to the original

## Original work

Based on [ResembleAI/chatterbox](https://huggingface.co/ResembleAI/chatterbox)
by Resemble AI, licensed under MIT. This derivative is also MIT-licensed.
---