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.

If you have any questions, feel free to contact us. (Emails: guanzzh.chen@gmail.com, lixin4ever@gmail.com)

Model Zoo

Model Name	Model Type	Starting Point	Train Data	Train Length	MAX Test Length	HF Repo
CLEX-LLaMA-2-7B-16K	base	LLaMA-2-7B	Redpajama-Book	16K	64K	link
CLEX-LLaMA-2-7B-Chat-16K	chat	CLEX-7B-16K	UltraChat	16K	64K	link
CLEX-LLaMA-2-7B-64K (this checkpoint)	base	LLaMA-2-7B	Redpajama-Book	64k	256K	link
CLEX-Phi-2-32K	base	Phi-2-2.7B	LongCorpus-2.5B	32k	128K	link
CLEX-Mixtral-8x7B-32K	base	Mixtral-8x7B-v0.1	LongCorpus-2.5B	32k	>128K	link
CLEX-Mixtral-8x7B-Chat-32k	chat	CLEX-Mixtral-8x7B-32K	Ultrachat 200k	32k	>128K	link

Usage

bash

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DAMO-NLP-SG/CLEX-LLaMA-2-7B-64K", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("DAMO-NLP-SG/CLEX-LLaMA-2-7B-64K", 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]))

Evaluation

Language Modelling

Here are the evaluation PPLs of the base models trained with CLEX. We apply training and evaluation on a subset of 2B tokens from the RedPajama-Book corpus, where the training and test sets are split by 99:1.

	Train Length	Eval.(32k)	Eval.(64k)	Eval.(128k)	Eval.(256k)
CLEX-LLaMA-2-7B	64k	5.99	5.89	6.04	5.98

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}
}