abhinavnmagic
commited on
Commit
•
3a3b32b
1
Parent(s):
2d1c7b6
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- fp8
|
4 |
+
- vllm
|
5 |
+
---
|
6 |
+
|
7 |
+
# Meta-Llama-3-70B-Instruct-FP8
|
8 |
+
|
9 |
+
## Model Overview
|
10 |
+
Meta-Llama-3-70B-Instruct quantized to FP8 weights and activations using per-tensor quantization, ready for inference with vLLM >= 0.5.0.
|
11 |
+
|
12 |
+
## Usage and Creation
|
13 |
+
Produced using [AutoFP8 with calibration samples from ultrachat](https://github.com/neuralmagic/AutoFP8/blob/147fa4d9e1a90ef8a93f96fc7d9c33056ddc017a/example_dataset.py).
|
14 |
+
|
15 |
+
## Evaluation
|
16 |
+
|
17 |
+
### Open LLM Leaderboard evaluation scores
|
18 |
+
| | Meta-Llama-3-70B-Instruct | Meta-Llama-3-70B-Instruct-FP8<br>(this model) |
|
19 |
+
| :------------------: | :----------------------: | :------------------------------------------------: |
|
20 |
+
| arc-c<br>25-shot | 72.69 | 72.61 |
|
21 |
+
| hellaswag<br>10-shot | 85.50 | 85.41 |
|
22 |
+
| mmlu<br>5-shot | 80.18 | 80.06 |
|
23 |
+
| truthfulqa<br>0-shot | 62.90 | 62.73 |
|
24 |
+
| winogrande<br>5-shot | 83.34 | 83.03 |
|
25 |
+
| gsm8k<br>5-shot | 92.49 | 91.12 |
|
26 |
+
| **Average<br>Accuracy** | **79.51** | **79.16** |
|
27 |
+
| **Recovery** | **100%** | **99.55%** |
|
28 |
+
|