# 8-bit quantization

ERNIE-4.5-21B-A3B-PT-8bit is an 8-bit quantized version of Baidu's ERNIE-4.5-21B-A3B-PT model, converted to MLX format and suitable for Apple Silicon devices.

Jan-nano-8bit is an 8-bit quantized version converted from the Menlo/Jan-nano model, optimized for the MLX framework and suitable for text generation tasks.

This is an 8-bit quantized version in MLX format converted from the DeepSeek-R1-0528-Qwen3-8B model, suitable for text generation tasks.

An 8-bit quantized version based on the DeepSeek-R1-0528-Qwen3-8B model, optimized for Apple Silicon chips and suitable for text generation tasks.

Devstral-Small-2505-8bit is an 8-bit quantized model converted from mistralai/Devstral-Small-2505, suitable for the MLX framework and supporting text generation tasks in multiple languages.

FastVLM-1.5B-Stage3-MNN is a text generation model based on the Transformer architecture. It is an 8-bit quantized version of FastVLM-1.5B-Stage3, suitable for text generation scenarios such as chatting.

This is a text-to-speech model based on the MLX format, supporting both English and Chinese, converted from prince-canuma/Spark-TTS-0.5B.

This is a text-to-speech model converted from sesame/csm-1b to MLX format, supporting the English language.

This model is an 8-bit quantized version converted from Qwen/Qwen3-235B-A22B, suitable for text generation tasks.

Orpheus is a high-performance Korean text-to-speech model, fine-tuned for natural emotional speech synthesis, offering an 8-bit quantized version for optimized efficiency.

Orpheus is a high-performance German text-to-speech model, fine-tuned to achieve natural and emotionally rich speech synthesis. This model is an 8-bit quantized version of the 3-billion-parameter model, optimized for operational efficiency.

Gemma 3 27B IT QAT 8bit is an MLX-format model converted from Google's Gemma 3 27B model, supporting image-to-text tasks.

This is an 8-bit quantized version of the Orpheus-3B pre-trained language model based on the MLX framework, originally developed by CanopyLabs.

The 8-bit quantized version of OmniGen-v1, suitable for text-to-image and image-to-image tasks, supporting multimodal input.

This is the 8-bit quantized version of EleutherAI's GPT-J 6B parameter model, optimized for running and fine-tuning on limited GPU resources (e.g., Colab or 1080Ti).