metadata
tags:
- fp8
- vllm
Meta-Llama-3-8B-Instruct-FP8
Model Overview
Meta-Llama-3-8B-Instruct quantized to FP8 weights and activations using per-tensor quantization, ready for inference with vLLM >= 0.5.0.
Usage and Creation
Produced using AutoFP8 with calibration samples from ultrachat.
Evaluation
Open LLM Leaderboard evaluation scores
| Meta-Llama-3-8B-Instruct | Meta-Llama-3-8B-Instruct-FP8 (this model) |
|
|---|---|---|
| arc-c 25-shot |
62.54 | 61.77 |
| hellaswag 10-shot |
78.83 | 78.56 |
| mmlu 5-shot |
66.60 | 66.27 |
| truthfulqa 0-shot |
52.44 | 52.35 |
| winogrande 5-shot |
75.93 | 76.4 |
| gsm8k 5-shot |
75.96 | 73.99 |
| Average Accuracy |
68.71 | 68.22 |
| Recovery | 100% | 99.28% |