Hugging Face's logo

BAAI
/
bge-reranker-v2.5-gemma2-lightweight

Text Classification
sentence-transformers
Safetensors
Transformers
multilingual
cost_wise_gemma
text-generation
custom_code
Model card Files
xet
 Community
13

How to setting the parameter to save flops? 如何设置节省flops的参数?

#1
by Kaguya-19 - opened
Discussion
Kaguya-19

Hello! bge-reranker-v2.5-gemma2-lightweight is a surprising model. How do we set the parameters like cutoff_layers to reproduce the results in the table?
你好！bge-reranker-v2.5-gemma2-轻量级是一个令人惊讶的模型。我们如何设置像cutoff_layers这样的参数来在表中重现结果呢?
cfli
 Beijing Academy of Artificial Intelligence org

Thanks for your attention. You just need to set compress_ratio as 2, compress_layer as [8] and cutoff_layers as [25].
Kaguya-19

Many Thanks
Kaguya-19 changed discussion status to closed
shizue
 Beijing Academy of Artificial Intelligence org

Hi, I want to ask if I don't set cutoff_layers and compression_{ratio, layer}, I will use the full model with 0% flops saving. Is that right?
cfli
 Beijing Academy of Artificial Intelligence org

Hi, I want to ask if I don't set cutoff_layers and compression_{ratio, layer}, I will use the full model with 0% flops saving. Is that right?

Yes.

Sign up or log in to comment