This model is a fine-tuned version of Qwen/Qwen2.5-32B-Instruct on the OpenThoughts-114k dataset dataset. This model improves upon the Bespoke-Stratos-32B model, which used 17k examples (Bespoke-Stratos-17k dataset).

Special thanks to the folks at Open Thoughts for fine-tuning this version of Qwen/Qwen2.5-32B-Instruct. More information about it can be found here:

https://huggingface.co/open-thoughts/OpenThinker-32B (Base Model)

https://github.com/open-thoughts/open-thoughts (Open Thoughts Git Repo)

I simply converted it to MLX format (using mlx-lm version 0.20.5.) with a quantization of 4-bit for better performance on Apple Silicon Macs.

Other Types:

Link	Type	Size	Notes
[MLX] (https://huggingface.co/AlejandroOlmedo/OpenThinker-32B-8bit-mlx)	8-bit	34.80 GB	Best Quality
[MLX] (https://huggingface.co/AlejandroOlmedo/OpenThinker-32B-4bit-mlx)	4-bit	18.40 GB	Good Quality

AlejandroOlmedo/OpenThinker-32B-4bit-mlx

The Model AlejandroOlmedo/OpenThinker-32B-4bit-mlx was converted to MLX format from open-thoughts/OpenThinker-32B using mlx-lm version 0.20.5.

Use with mlx

pip install mlx-lm

python

from mlx_lm import load, generate

model, tokenizer = load("AlejandroOlmedo/OpenThinker-32B-4bit-mlx")

prompt="hello"

if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

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