cs2764's picture
Upload MLX converted model with quantization settings
5b6a1f9 verified
metadata
license: other
license_name: modified-mit
library_name: transformers
base_model: moonshotai/Kimi-K2-Instruct-0905
tags:
  - mlx

cs2764/Kimi-K2-Instruct-0905-mlx-mixed_3_4

The Model cs2764/Kimi-K2-Instruct-0905-mlx-mixed_3_4 was converted to MLX format from moonshotai/Kimi-K2-Instruct-0905 using mlx-lm version 0.28.0.

Quantization Details

This model was converted with the following quantization settings:

  • Quantization Strategy: mixed_3_4 (Mixed precision)
  • Average bits per weight: 3.663

Use with mlx

pip install mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("cs2764/Kimi-K2-Instruct-0905-mlx-mixed_3_4")

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)