|
--- |
|
license: apache-2.0 |
|
--- |
|
|
|
--- |
|
|
|
**Paper**: [https://arxiv.org/pdf/2310.06694.pdf](https://arxiv.org/pdf/2310.06694.pdf) |
|
**Code**: https://github.com/princeton-nlp/LLM-Shearing |
|
**Models**: [Sheared-LLaMA-1.3B](https://huggingface.co/princeton-nlp/Sheared-LLaMA-1.3B), [Sheared-LLaMA-2.7B](https://huggingface.co/princeton-nlp/Sheared-LLaMA-2.7B) |
|
|
|
**License**: Must comply with license of Llama2 since it's a model derived from Llama2. |
|
|
|
--- |
|
|
|
Sheared-LLaMA-2.7B is a model pruned and further pre-trained from [meta-llama/Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf). We dynamically load data from different domains in the [RedPajama dataset](https://github.com/togethercomputeub.com/togethercomputer/RedPajama-Data). We use 0.4B tokens for pruning and 50B tokens for continued pre-training the pruned model. This model can be loaded into huggingface via |
|
|
|
``` |
|
model = AutoModelForCausalLM.from_pretrained("princeton-nlp/Sheared-LLaMA-2.7B") |
|
``` |
|
|
|
- Smaller-scale |
|
- Same vocabulary as LLaMA1 and LLaMA2 |
|
- Derived with a budget of 50B tokens by utilizing existing strong LLMs |
|
|
|
## Downstream Tasks |
|
|
|
We evaluate on an extensive set of downstream tasks including reasoning, reading comprehension, language modeling and knowledge intensive tasks. Our Sheared-LLaMA models outperform existing large language models. |
|
|
|
| Model | # Pre-training Tokens | Average Performance | |
|
| ------------------- | --------------------- | ------------------- | |
|
| LLaMA2-7B | 2T | 64.6 | |
|
|
|
**1.3B** |
|
|
|
| Model | # Pre-training Tokens | Average Performance | |
|
| ------------------- | --------------------- | ------------------- | |
|
| OPT-1.3B | 300B | 48.2 | |
|
| Pythia-1.4B | 300B | 48.9 | |
|
| Sheared-LLaMA-1.3B | 50B | 51.0 | |
|
|
|
**3B** |
|
|
|
| Model | # Pre-training Tokens | Average Performance | |
|
| ------------------- | --------------------- | ------------------- | |
|
| OPT-2.7B | 300B | 51.4 | |
|
| Pythia-2.8B | 300B | 52.5 | |
|
| INCITE-Base-3B | 800B | 54.7 | |
|
| Open-LLaMA-3B-v1 | 1T | 55.1 | |
|
| Open-LLaMA-3B-v2 | 1T | 55.7 | |
|
| **Sheared-LLaMA-2.7B** | **50B** | **56.7** | |
|
|
|
## Bibtex |
|
``` |
|
@article{xia2023sheared, |
|
title={Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning}, |
|
author={Xia, Mengzhou and Gao, Tianyu, and Zeng, Zhiyuan and Chen, Danqi}, |
|
year={2023} |
|
} |
|
``` |
|
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) |
|
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_princeton-nlp__Sheared-LLaMA-2.7B) |
|
|
|
| Metric | Value | |
|
|-----------------------|---------------------------| |
|
| Avg. | 35.66 | |
|
| ARC (25-shot) | 41.72 | |
|
| HellaSwag (10-shot) | 71.01 | |
|
| MMLU (5-shot) | 26.92 | |
|
| TruthfulQA (0-shot) | 37.32 | |
|
| Winogrande (5-shot) | 67.01 | |
|
| GSM8K (5-shot) | 1.06 | |
|
| DROP (3-shot) | 4.57 | |
|
|
