Qwen3-Swallow-32B-RL-v0.2-MLX-fp16

This model is an MLX format conversion of tokyotech-llm/Qwen3-Swallow-32B-RL-v0.2, optimized for Apple Silicon.

Model Details

Attribute	Value
Original Model	tokyotech-llm/Qwen3-Swallow-32B-RL-v0.2
Architecture	Dense Transformer
Parameters	32B
Quantization	Full precision (fp16)
Model Size	~61 GB
Format	MLX (Apple Silicon optimized)
Converted with	mlx-lm v0.30.8
License	Apache 2.0

About Qwen3-Swallow

Qwen3-Swallow is a bilingual Japanese-English large language model developed by the Swallow Project at the Institute of Science Tokyo (formerly Tokyo Institute of Technology) and AIST. Built upon Qwen3 through Continual Pre-Training (CPT), Supervised Fine-Tuning (SFT), and Reinforcement Learning (RL), it achieves strong performance on both Japanese and English tasks while maintaining capabilities in mathematics and coding.

For more details, see the original model card.

Usage

Quick Start (Python)

from mlx_lm import load, generate

model, tokenizer = load("tocchitocchi/Qwen3-Swallow-32B-RL-v0.2-MLX-fp16")

messages = [{"role": "user", "content": "hello"}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

response = generate(model, tokenizer, prompt=prompt, verbose=True, max_tokens=512)

Interactive Chat

mlx_lm.chat --model tocchitocchi/Qwen3-Swallow-32B-RL-v0.2-MLX-fp16

OpenAI-Compatible Server

mlx_lm.server --model tocchitocchi/Qwen3-Swallow-32B-RL-v0.2-MLX-fp16 --port 8080

Then connect with any OpenAI-compatible client at http://localhost:8080/v1.

Acknowledgments

• Original model by Swallow Project (Institute of Science Tokyo and AIST)
• MLX framework by Apple Machine Learning Research
• Conversion performed using mlx-lm