Sparse-Llama-3.1-8B-gsm8k-2of4-quantized.w4a16

Model Overview

• Model Architecture: Llama-3.1-8B
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Sparsity: 2:4
  • Weight quantization: INT4
• Release Date: 11/21/2024
• Version: 1.0
• License(s): llama3.1
• Model Developers: Neural Magic

This is AI model especialized in grade-school math obtained by fine-tuning the 2:4 sparse Sparse-Llama-3.1-8B-2of4 on the GSM8k dataset, followed by one-shot quantization. It achieves 64.3% 0-shot accuracy on the test set of GSM8k, compared to 66.3% for the fine-tuned dense model Llama-3.1-8B-gsm8k — demonstrating over 96.9% accuracy recovery. In constrast, the pretrained Llama-3.1-8B achieves 50.7% 5-shot accuracy and the sparse foundational Sparse-Llama-3.1-8B-2of4 model achieves 56.3% 5-shot accuracy.

Model Optimizations

This model was obtained by quantizing the weights of Sparse-Llama-3.1-8B-gsm8k-2of4 to INT4 data type. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%. That is on top of the reduction of 50% of weights via 2:4 pruning employed on Sparse-Llama-3.1-8B-gsm8k-2of4.

Only the weights of the linear operators within transformers blocks are quantized. Symmetric per-channel quantization is applied, in which a linear scaling per output dimension maps the INT4 and floating point representations of the quantized weights. The GPTQ algorithm is applied for quantization, as implemented in the llm-compressor library.

Deployment with vLLM

This model can be deployed efficiently using the vLLM backend. vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Evaluation

This model was evaluated on the lm-evaluation-harness.

Accuracy

GSM8k Benchmark

Metric	Llama-3.1-8B (5-shot)	Sparse-Llama-3.1-8B-2of4 (5-shot)	Llama-3.1-8B-gsm8k (0-shot)	Sparse-Llama-3.1-8B-gsm8k-2of4 (0-shot)	Sparse-Llama-3.1-8B-gsm8k-2of4-quantized.w4a16 (0-shot)
Accuracy	50.7%	56.3%	66.3%	66.9%	64.3%