L

Llama 2 7b Hf 4bit 64rank

Developed by LoftQ
The LoftQ (LoRA Fine-tuning Aware Quantization) model provides a quantized backbone network and LoRA adapters, specifically designed for LoRA fine-tuning to improve the fine-tuning performance and efficiency of large language models during the quantization process.
Downloads 1,754
Release Time : 11/21/2023

Model Overview

This model is based on LLAMA-2-7b and is 4-bit quantized using the LoftQ method. It also provides LoRA adapters to address the incompatibility issue between large language models and LoRA fine-tuning during the quantization process.

Model Features

Quantization Support
Provides a 4-bit quantized backbone network, significantly reducing the model storage and computational resource requirements.
LoRA Fine-tuning Awareness
A quantization method specifically designed for LoRA fine-tuning to optimize the performance and efficiency during the fine-tuning process.
Efficient Storage
The quantized model size is approximately 4.2 GiB, suitable for resource-constrained environments.

Model Capabilities

Text Generation
LoRA Fine-tuning

Use Cases

Mathematical Problem Solving
GSM8K Mathematical Problem Solving
After fine-tuning on the GSM8K dataset, the model can be used to solve mathematical problems.
The accuracy of the fine-tuned model on GSM8K is 35.0%.
Text Generation
WikiText-2 Text Generation
Fine-tuned on the WikiText-2 dataset for generating coherent text.
The perplexity of the fine-tuned model on WikiText-2 is 5.24.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase