search

liuqi6777 / pe_rank

Leveraging passage embeddings for efficient listwise reranking with large language models.
Apache License 2.0
33 stars 2 forks source link

training loss #4

Open Jason-kid opened 2 months ago

Jason-kid commented 2 months ago

bother you again。 what is the loss in two stage in the experiments? with my own coupus and chinese llm, the loss for alignment stage is 1.5; the learn to rank is still training with now loss=60.

liuqi6777 commented 2 months ago

hi, the loss for the alignment stage is around 1-2, and the loss for the ltr stage is much higher and may vary depending on the model and dataset, but in my experience, tens are normal values.