efrainmain/draft-chunkformer-finetune-encoder

1# ChunkFormer fine-tuned checkpoint2 3This repo contains:4 5- `pytorch_model.bin`: best model weights for inference / `ChunkFormerModel.from_pretrained`.6- `resume_checkpoint/training_checkpoint.pt`: latest full checkpoint for resuming training.7- `resume_checkpoint/model_state_dict.pt`: latest model state dict only.8- `resume_checkpoint/optimizer.pt`: latest optimizer state, if available.9- `resume_checkpoint/scheduler.pt`: latest scheduler state, if available.10- `resume_checkpoint/scaler.pt`: latest AMP GradScaler state, if available.11- `best_checkpoints.json`: validation loss metadata.12- `history.json`: training history.13 14Latest trained epoch: 215Latest train_loss: 178.6814606852464416Latest val_loss: 155.4572542757601417 18Best epoch: 219Best train_loss: 178.6814606852464420Best val_loss: 155.4572542757601421 22Base model: khanhld/chunkformer-ctc-large-vie23