File size: 684 Bytes
09c90f3 4c0d06b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 |
---
library_name: transformers
tags:
- llama
- facebook
- meta
- llama-3
- conversational
- text-generation-inference
---
Official [AQLM](https://arxiv.org/abs/2401.06118) quantization of [meta-llama/Meta-Llama-3-70B
](https://huggingface.co/meta-llama/Meta-Llama-3-70B).
For this quantization, we used 1 codebook of 16 bits.
Results (in progress):
| Model | Quantization | MMLU (5-shot) | ArcC| ArcE| Hellaswag | Winogrande | PiQA | Model size, Gb |
|------|------|-------|------|------|------|------|------|------|
|meta-llama/Meta-Llama-3-70B | - | 0.6596 | 0.8935 | 0.6921 | 0.8444 | 0.8090 | 141.2 |
| | 1x16 | 0.7475| 0.6049 | 0.8468 | 0.6520 | 0.8248 | 0.7790 | 21.9 | |