CLEX: Continuous Length Extrapolation for Large Language Models
This repo stores the checkpoint of CLEX-7B-Chat-16K
Features and Highlights of CLEX
- Simple and Clear: MINIMAL code and architecture changes. Only one up-and-down projection layer introduced, NO recurrent memory caching or sparse attention required.
- Train Short, Test Long: NO performance drop on the sequences 4x~8x longer than the training ones (see here).
- Continuous Length Extrapolation: Explicitly modeling the continuous dynamics of context window size during length extrapolation.
More details about long-text modeling with our CLEX can be found at the git repo.
Model Zoo
Model Name | Model Type | Starting Point | Train Data | Train Length | MAX Test Length |
---|---|---|---|---|---|
CLEX-7B-4K | base | LLaMA-2-7B | Redpajama-Book | 4K | 16K |
CLEX-7B-Chat-4K | chat | CLEX-7B-4K | UltraChat | 4K | 16K |
CLEX-7B-16K | base | LLaMA-2-7B | Redpajama-Book | 16K | 64K |
CLEX-7B-Chat-16K (this checkpoint) | chat | CLEX-7B-16K | UltraChat | 16K | 64K |
How to Use
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("DAMO-NLP-SG/CLEX-7B-Chat-16K", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("DAMO-NLP-SG/CLEX-7B-Chat-16K", torch_dtype=torch.bfloat16, trust_remote_code=True)
inputs = tokenizer("What is CLEX?", return_tensors="pt")
sample = model.generate(**inputs, max_length=128)
print(tokenizer.decode(sample[0]))
Citation
If you find our project useful, hope you can star our repo and cite our paper as follows:
@article{damonlpsg2023clex,
author = {Chen, Guanzheng and Li, Xin and Meng, Zaiqiao and Liang, Shangsong and Bing, Lidong},
title = {CLEX: Continuous Length Extrapolation for Large Language Models},
year = 2023,
journal = {arXiv preprint arXiv:2310.16450},
url = {https://arxiv.org/abs/2310.16450}
}
- Downloads last month
- 40
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.