metadata
base_model: MiniMaxAI/MiniMax-M2.5
language: en
library_name: mlx-lm
license: apache-2.0
model_name: MiniMax-M2.5-mix3-6bit
tags:
- quantization
- mixed_3_6
- minimax
- mlx
base_model: MiniMaxAI/MiniMax-M2.5 language: en library_name: mlx-lm license: modified-mit model_name: MiniMax-M2.5-mix3-6bit tags: - quantization - mixed_3_6 - minimax - mlx
MiniMax-M2.5-mix3-6bit
Mixed precision quantized version of MiniMax M2.5 using mlx-lm with --quant-predicate mixed_3_6.
Model Details
| Property | Value |
|---|---|
| Base Model | MiniMaxAI/MiniMax-M2.5 |
| Quantization | mlx-lm v0.30.7 with --quant-predicate mixed_3_6 |
| Library | mlx-lm |
| License | modified-mit |
Inference Parameters
| Parameter | Value |
|---|---|
| temperature | 1.0 |
| top_p | 0.95 |
| top_k | 40 |
Usage
import mlx_lm
from mlx_lm.sample_utils import make_sampler
model_path = "petergilani/MiniMax-M2.5-mix3-6bit"
model, tokenizer = mlx_lm.load(model_path)
sampler = make_sampler(temp=1.0, top_p=0.95, top_k=40)
prompt = "Your prompt here"
response = mlx_lm.generate(
model,
tokenizer,
prompt=prompt,
sampler=sampler,
max_tokens=512
)
print(response)