huggingface / nanotron

Minimalistic large language model 3D-parallelism training
Apache License 2.0
1.14k stars 107 forks source link

We don't save checkpoint after training ends #163

Open NouamaneTazi opened 4 months ago

NouamaneTazi commented 4 months ago

After the end of training we don't save the model if it's not a multiple of checkpoint_interval in config. We'd like to have an option in config to save latest checkpoint too like we do for save_initial_state