search

James-QiuHaoran / LLM-serving-with-proxy-models

Efficient Interactive LLM Serving with Proxy Model-based Sequence Length Prediction
Apache License 2.0
17 stars 5 forks source link

Add support for importing new training dataset in addition to LMSYS-Chat-1M #7

Closed James-QiuHaoran closed 4 months ago

James-QiuHaoran commented 4 months ago

Currently, the predictor is trained on the LMSYS-Chat-1M dataset. Support and some interface should be added to allow training on new datasets

James-QiuHaoran commented 4 months ago

Added support for training the predictor with a customized dataset in addition to LMSYS dataset: https://github.com/James-QiuHaoran/LLM-serving-with-proxy-models/commit/0089a63519293914cf661189aa749ee3d493cacf

An example use case can be found in the README.