4-bit Quantized Llama 3 Model

Description

This repository hosts the 4-bit quantized version of the Llama 3 model. Optimized for reduced memory usage and faster inference, this model is suitable for deployment in environments where computational resources are limited.

Model Details

Model Type: Transformer-based language model.
Quantization: 4-bit precision.
Advantages:
- Memory Efficiency: Reduces memory usage significantly, allowing deployment on devices with limited RAM.
- Inference Speed: Accelerates inference times, depending on the hardware's ability to process low-bit computations.

How to Use

To utilize this model efficiently, follow the steps below:

Loading the Quantized Model

Load the model with specific parameters to ensure it utilizes 4-bit precision:

from transformers import AutoModelForCausalLM

model_4bit = AutoModelForCausalLM.from_pretrained("SweatyCrayfish/llama-3-8b-quantized", device_map="auto", load_in_4bit=True)

Adjusting Precision of Components

Adjust the precision of other components, which are by default converted to torch.float16:

import torch
from transformers import AutoModelForCausalLM

model_4bit = AutoModelForCausalLM.from_pretrained("SweatyCrayfish/llama-3-8b-quantized", load_in_4bit=True, torch_dtype=torch.float32)
print(model_4bit.model.decoder.layers[-1].final_layer_norm.weight.dtype)

Citation

Original repository and citations: @article{llama3modelcard, title={Llama 3 Model Card}, author={AI@Meta}, year={2024}, url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md} }