Skip to content

apd10/USA_InfLLM

 
 

Repository files navigation

Built on the top of InfLLM repo

python benchmark/pred.py --config_path llama.yaml --output_dir_path benchmark/test/ --datasets multifieldqa_zh \
                          --load_usa artifacts/usa.all25.t64.b32.th0.1400itr.pt --chunk_size 256 --max_prompt_len 19000 \
                          --prefetch_offset 128 --token_budget 256 --limit 100 --baseline usa

About

The code of our paper "InfLLM: Unveiling the Intrinsic Capacity of LLMs for Understanding Extremely Long Sequences with Training-Free Memory"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 99.2%
  • Shell 0.8%