rojasdiego/Meta-Llama-3.1-8B-Instruct-Apple-MLX-Adapter

Meta-Llama-3.1-8B-Instruct-Apple-MLX

Overview

This model is a fine-tuned version of Meta's LLaMa 3.1 8B model, specifically adapted to answer questions and provide guidance on Apple's latest machine learning framework, MLX. The fine-tuning was done using the LORA (Low-Rank Adaptation) method on a custom dataset of question-answer pairs derived from the MLX documentation.

Dataset

Fine-tuned on a single epoch of Apple MLX QA.

Installation

To use the model, you need to install the required dependencies:

pip install peft transformers jinja2==3.1.0

Usage

Here’s a sample code snippet to load and interact with the model:

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the base model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.1-8B-Instruct", torch_dtype=torch.bfloat16
)
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")

# Load the fine-tuned model using LORA
model = PeftModel.from_pretrained(
    model,
    "rojas-diego/Meta-Llama-3.1-8B-Instruct-Apple-MLX",
).to("cuda")

# Define input using a chat template with a system prompt and user query
ids = tokenizer.apply_chat_template(
    [
        {
            "role": "system",
            "content": "You are a helpful AI coding assistant with expert knowledge of Apple's latest machine learning framework: MLX. You can help answer questions about MLX, provide code snippets, and help debug code.",
        },
        {
            "role": "user",
            "content": "How do you transpose a matrix in MLX?",
        },
    ],
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt",
).to("cuda")

# Generate and print the response
print(
    tokenizer.decode(
        model.generate(input_ids=ids, max_new_tokens=256, temperature=0.5).tolist()[0][
            len(ids) :
        ]
    )
)