---
license: apache-2.0
base_model:
  - Kwaipilot/KAT-Coder-V2.5-Dev
pipeline_tag: text-generation
tags:
- mlx
---

# KAT-Coder-V2.5-Dev-oQ3e

A 3-bit enhanced oQ quantization of
[Kwaipilot/KAT-Coder-V2.5-Dev](https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev)
for Apple Silicon and MLX-compatible runtimes.

## Quantization details

The full calibration summary is available in
[`oq_imatrix_report.json`](./oq_imatrix_report.json).

**Note:** This quant was created using **float16** non-quant weights. Per oMLX, float16 gives ~20% faster prefill on M1/M2 Apple Silicon (native fp16). bfloat16 is safer on M3+ and for numerical stability. For the standard oQe with bfloat16, see [this repo](https://huggingface.co/ZQ-Dev/KAT-Coder-V2.5-Dev-oQ3e).

## License
Apache 2.0, inherited from the base model.