Gausson/

pythia-160m-deduped-n64h-SepLLM

CUDA_LAUNCH_BLOCKING=1
lm_eval --model hf \
	--model_args pretrained=Gausson/pythia-160m-deduped-n64h-SepLLM \
	--tasks  arc_challenge,arc_easy,lambada_openai,logiqa,piqa,sciq,winogrande,wsc,wikitext  \
	--num_fewshot 5 \
	--device cuda:0\
	--batch_size 32

hf (pretrained=Gausson/pythia-160m-deduped-n64h-SepLLM), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: 32
|    Tasks     |Version|Filter|n-shot|    Metric     |   | Value |   |Stderr|
|--------------|------:|------|-----:|---------------|---|------:|---|------|
|arc_challenge |      1|none  |     5|acc            |↑  | 0.2073|±  |0.0118|
|              |       |none  |     5|acc_norm       |↑  | 0.2432|±  |0.0125|
|arc_easy      |      1|none  |     5|acc            |↑  | 0.4844|±  |0.0103|
|              |       |none  |     5|acc_norm       |↑  | 0.4516|±  |0.0102|
|lambada_openai|      1|none  |     5|acc            |↑  | 0.3047|±  |0.0064|
|              |       |none  |     5|perplexity     |↓  |36.5298|±  |1.2471|
|logiqa        |      1|none  |     5|acc            |↑  | 0.2535|±  |0.0171|
|              |       |none  |     5|acc_norm       |↑  | 0.2596|±  |0.0172|
|piqa          |      1|none  |     5|acc            |↑  | 0.6442|±  |0.0112|
|              |       |none  |     5|acc_norm       |↑  | 0.6398|±  |0.0112|
|sciq          |      1|none  |     5|acc            |↑  | 0.8060|±  |0.0125|
|              |       |none  |     5|acc_norm       |↑  | 0.7910|±  |0.0129|
|wikitext      |      2|none  |     5|bits_per_byte  |↓  | 0.9217|±  |   N/A|
|              |       |none  |     5|byte_perplexity|↓  | 1.8943|±  |   N/A|
|              |       |none  |     5|word_perplexity|↓  |30.4533|±  |   N/A|
|winogrande    |      1|none  |     5|acc            |↑  | 0.5146|±  |0.0140|
|wsc           |      1|none  |     5|acc            |↑  | 0.3846|±  |0.0479|

@inproceedings{chen2025sepllm,
  title={{SepLLM: Accelerate Large Language Models by Compressing One Segment into One Separator}},
  author={Chen, Guoxuan and Shi, Han and Li, Jiawei and Gao, Yihang and Ren, Xiaozhe and Chen, Yimeng and Jiang, Xin and Li, Zhenguo and Liu, Weiyang and Huang, Chao},
  booktitle={International Conference on Machine Learning},
  year={2025},
  note={Also available at arXiv:2412.12094}
}