End of training

README.md

@@ -46,7 +46,7 @@ More information needed
 
 # Resource Usage Comparison
 
-- VRAM Use: 7.7871 GB
+- VRAM Use: 7.7868 GB
 
 # Distillation (Teacher -> Student) Architecture Difference:
 
@@ -66,7 +66,7 @@ More information needed
 <br/>
 
 # Train Dataset
-Trained on 145,744,973 tokens from the [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dataset.
+Trained on 145,714,513 tokens from the [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dataset.
 
 - Num Samples: `247,500`
 - Subset: `20231101.en`
@@ -76,7 +76,7 @@ Trained on 145,744,973 tokens from the [wikimedia/wikipedia](https://huggingface
 # Training Objective
 
 ```
-DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=25.0, loss_fn=raw_mse, layer_mapper=layer-2, norm=layernorm, projector=orthogonal))
+DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=25.0, loss_fn=cos, layer_mapper=layer-2, norm=batchnorm, projector=orthogonal))
 ```
 
 # Hyperparameters
@@ -93,9 +93,9 @@ The following hyperparameters were used during training:
 - lr_scheduler_type: `cosine_with_min_lr`
 - lr_scheduler_warmup_ratio: `0.5`
 - num_epochs: `1.0`
-- distillation_objective: `DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=25.0, loss_fn=raw_mse, layer_mapper=layer-2, norm=layernorm, projector=orthogonal))`
+- distillation_objective: `DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=25.0, loss_fn=cos, layer_mapper=layer-2, norm=batchnorm, projector=orthogonal))`
 - train_embeddings: `True`
-- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7fe929482740>`
+- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7fc4addc8820>`
 - student_model_name_or_path: `None`
 - student_config_name_or_path: `None`
 - student_model_config: `None`
@@ -114,7 +114,7 @@ The following hyperparameters were used during training:
 - dataset_test_size: `0.01`
 - gradient_accumulation_steps: `1`
 - weight_decay: `0.0`
-- max_grad_norm: `1.0`
+- max_grad_norm: `100`
 - warmup_ratio: `0.5`
 - warmup_steps: `0`
 - gradient_checkpointing: `True`