|
--- |
|
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-3-8B-Instruct |
|
](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct). |
|
|
|
For this quantization, we used 1 codebook of 16 bits. |
|
|
|
Results: |
|
| Model | Quantization | MMLU (5-shot) | GSM8k (8-shot) | ArcC| ArcE| Hellaswag | Winogrande | PiQA | Model size, Gb | |
|
|------|------|-------|------|------|------|------|------|------|------| |
|
|meta-llama/Meta-Llama-3-8B-Instruct| None |0.6560 | 0.7475 | 0.5299 | 0.8165 | 0.5771 | 0.7867 | 0.7206 | 16.1 | |
|
| | 1x16 | 0.5529 | 0.3397 | 0.4428 | 0.7580 | 0.5313 | 0.7666 | 0.6788 | 4.1 | |
