File size: 3,410 Bytes
9bb31e7 bac07a9 9bb31e7 bac07a9 9bb31e7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 |
---
license: mit
---
# Model Card for SciLitLLM1.5
SciLitLLM1.5 adapts a general large language model for effective scientific literature understanding. Starting from the Qwen2.5-7B/14B model, SciLitLLM1.5-7B/14B goes through a hybrid strategy that integrates continual pre-training (CPT) and supervised fine-tuning (SFT), to simultaneously infuse scientific domain knowledge and enhance instruction-following capabilities for domain-specific tasks.
In this process, we identify two key challenges: (1) constructing high-quality CPT corpora, and (2) generating diverse SFT instructions. We address these challenges through a meticulous pipeline, including PDF text extraction, parsing content error correction, quality filtering, and synthetic instruction creation.
Applying this strategy, we present SciLitLLM-7B and 14B, specialized in scientific literature understanding, which demonstrates promising performance on scientific literature understanding benchmarks.
We observe promising performance enhancements, **with an average improvement of 4.0\% on SciAssess and 10.1% on SciRIFF, compared to the leading LLMs under 10B parameters**. Notably, **SciLitLLM-7B even outperforms Llama3.1 and Qwen2.5 with 70B parameters on SciRIFF**. Additionally, SciLitLLM-14B achieves leading results on both benchmarks, surpassing other open-source LLMs. Further ablation studies demonstrate the effectiveness of each module in our pipeline.
See the [paper](https://arxiv.org/abs/2408.15545) for more details and [github](https://github.com/dptech-corp/Uni-SMART) for data processing codes.
## Requirements
Since SciLitLLM is based on Qwen2.5, we advise you to install `transformers>=4.37.0`, or you might encounter the following error:
```
KeyError: 'qwen2'
```
## Quickstart
Here provides a code snippet with `apply_chat_template` to show you how to load the tokenizer and model and how to generate contents.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained(
"Uni-SMART/SciLitLLM1.5-7B",
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Uni-SMART/SciLitLLM1.5-7B")
prompt = "Can you summarize this article for me?\n <ARTICLE>"
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```
## Citation
If you find our work helpful, feel free to give us a cite.
```
@misc{li2024scilitllmadaptllmsscientific,
title={SciLitLLM: How to Adapt LLMs for Scientific Literature Understanding},
author={Sihang Li and Jin Huang and Jiaxi Zhuang and Yaorui Shi and Xiaochen Cai and Mingjun Xu and Xiang Wang and Linfeng Zhang and Guolin Ke and Hengxing Cai},
year={2024},
eprint={2408.15545},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2408.15545},
}
``` |