catalystsec/MiniMax-M2.7-4bit-DWQ

This model was quantized to 4-bit using DWQ with mlx-lm version 0.31.2.

Parameter	Value
DWQ learning rate	2e-7
Batch size	1
Dataset	allenai/tulu-3-sft-mixture
Initial validation loss	0.051
Final validation loss	0.032
Relative KL reduction	≈37 %
Tokens processed	≈1.16 M

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("catalystsec/MiniMax-M2.7-4bit-DWQ")

prompt = "hello"

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

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