search

THUDM / AgentTuning

AgentTuning: Enabling Generalized Agent Abilities for LLMs
https://thudm.github.io/AgentTuning/
1.36k stars 95 forks source link

Number of training steps #36

Closed Mayer123 closed 12 months ago

Mayer123 commented 12 months ago

Great work for creating the AgentTuning data! I'm wondering how many training steps did you use for training the final AgentLM-7B model?

Btlmd commented 12 months ago

Sorry for not mentioning this in the appendix. AgentLM-7B is trained for 2k steps. Note that packing is used in training, so the number of instances learned in a batch is more than the value of batch size.