---
tags:
- fp8
- vllm
---

# Meta-Llama-3-70B-Instruct-FP8

## Model Overview
Meta-Llama-3-70B-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](https://github.com/neuralmagic/AutoFP8/blob/147fa4d9e1a90ef8a93f96fc7d9c33056ddc017a/example_dataset.py).

## Evaluation

### Open LLM Leaderboard evaluation scores
|                      | Meta-Llama-3-70B-Instruct | Meta-Llama-3-70B-Instruct-FP8<br>(this model) |
| :------------------: | :----------------------: | :------------------------------------------------: |
| arc-c<br>25-shot     | 72.69                    | 72.61                                              |
| hellaswag<br>10-shot | 85.50                    | 85.41                                              |
| mmlu<br>5-shot       | 80.18                    | 80.06                                              |
| truthfulqa<br>0-shot | 62.90                    | 62.73                                              |
| winogrande<br>5-shot | 83.34                    | 83.03                                              |
| gsm8k<br>5-shot      | 92.49                    | 91.12                                              |
| **Average<br>Accuracy**  | **79.51**                    |              **79.16**                                     |
| **Recovery**             | **100%**                     |              **99.55%**                                     |