|
--- |
|
library_name: transformers |
|
tags: |
|
- aqlm |
|
- llama |
|
- facebook |
|
- meta |
|
- llama-3 |
|
- conversational |
|
- text-generation-inference |
|
base_model: meta-llama/Llama-3.2-3B-Instruct |
|
--- |
|
|
|
Official [AQLM](https://arxiv.org/abs/2401.06118) quantization of [meta-llama/Llama-3.2-3B |
|
](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct) finetuned with [PV-Tuning](https://arxiv.org/abs/2405.14852). |
|
|
|
For this quantization, we used 2 codebooks of 8 bits and groupsize of 8. |
|
|
|
Results: |
|
| Model | Quantization | MMLU (5-shot) | ArcC | ArcE | Hellaswag | PiQA | Winogrande | Model size, Gb | |
|
|-------|--------------|---------------|--------|--------|-----------|--------|------------|----------------| |
|
| meta-llama/Llama-3.2-3B-Instruct | fp16 | 0.5984 | 0.4369 | 0.7428 | 0.5224 | 0.7579 | 0.6732 | 6.4 | |
|
| | 2x8g8 | 0.4842 | 0.3686 | 0.7066 | 0.4833 | 0.7274 | 0.6346 | 1.5 | |
|
|
