|
--- |
|
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-Instruct |
|
](https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct). |
|
|
|
For this quantization, we used 1 codebook of 16 bits. |
|
|
|
Results (measured with `lm_eval==4.0`): |
|
| Model | Quantization | MMLU (5-shot) | ArcC| ArcE| Hellaswag | Winogrande | PiQA | Model size, Gb | |
|
|------|------|-------|------|------|------|------|------|------| |
|
|meta-llama/Meta-Llama-3-70B | - | 0.7980 | 0.6160 | 0.8624 | 0.6367 | 0.8183 | 0.7632 | 141.2 | |
|
| | 1x16 | 0.7587 | 0.4863 | 0.7668 | 0.6159 | 0.7481 | 0.7537 | 21.9 | |
