Phi-3.5-vision-mlx-int4

This is a quantized INT4 model based on Apple MLX Framework Phi-3.5-vision-instruct. You can deploy it on Apple Silicon devices (M1,M2,M3).

Note: This is unoffical version,just for test and dev.

Installation

pip install -U mlx-vlm

Conversion

python -m mlxv_lm.convert --hf-path microsoft/Phi-3.5-vision-instruct -q

Samples

import mlx.core as mx
from mlx_vlm import load, generate

model_path = "Your Phi-3.5-vision-mlx-int4 Path"
model, processor = load(model_path,processor_config={"trust_remote_code":"True"})

messages = [
                {"role": "user", "content": "<|image|>\n Summarize the pic"}, 
]

prompt = processor.tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

output = generate(model, processor, "Your Image Path", placeholder+prompt, verbose=False, max_tokens=1024)

output