Training in progress, step 198000

README.md

@@ -44,7 +44,7 @@ More information needed
 
 # Resource Usage Comparison
 
-- VRAM Use: 7.4164 GB
+- VRAM Use: 7.4146 GB
 
 # Distillation (Teacher -> Student) Architecture Difference:
 
@@ -85,7 +85,7 @@ Trained on 226,096,614 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=5, loss_fn=raw_mse, layer_mapper=layer-2, projector=orthogonal))
+DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=25, loss_fn=raw_mse, layer_mapper=layer-2, projector=mlp))
 ```
 
 # Hyperparameters
@@ -94,17 +94,16 @@ The following hyperparameters were used during training:
 <details>
 <summary>Expand</summary>
 
-- learning_rate: `0.0001`
+- learning_rate: `0.0002`
 - train_batch_size: `4`
 - eval_batch_size: `8`
 - seed: `42`
 - optimizer: `Adam with betas=(0.9,0.999) and epsilon=1e-08`
 - lr_scheduler_type: `polynomial`
-- lr_scheduler_warmup_ratio: `0.2`
 - 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=5, loss_fn=raw_mse, layer_mapper=layer-2, projector=orthogonal))`
+- distillation_objective: `DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=25, loss_fn=raw_mse, layer_mapper=layer-2, projector=mlp))`
 - train_embeddings: `True`
-- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7f0234639960>`
+- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7fcf483061a0>`
 - student_model_name_or_path: `None`
 - student_config_name_or_path: `distilbert/distilgpt2`
 - student_model_config: `None`
@@ -124,7 +123,7 @@ The following hyperparameters were used during training:
 - gradient_accumulation_steps: `1`
 - weight_decay: `0.0`
 - max_grad_norm: `1.0`
-- warmup_ratio: `0.2`
+- warmup_ratio: `0`
 - warmup_steps: `0`
 - gradient_checkpointing: `True`
 
@@ -135,5 +134,5 @@ The following hyperparameters were used during training:
 # Framework Versions
 - Distily 0.4.1
 - Transformers 4.44.2
-- Pytorch 2.4.0+cu121
+- Pytorch 2.3.0
 - Datasets 2.21.0

logs/attn_norm=layernorm, attn_projector=mlp, attn_weight=5, learning_rate=0.0001, per_device_train_batch_size=2, warmup_ratio=0/events.out.tfevents.1725074707.63bdd91cff03

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64036b223b1d5450e2a2e977785d9da52ebf3bff28119135cb0b1708a73bd53a
+size 95114820

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55045d7f189faab391137497d4ceb55cabf51b994261207780438aac3a775aba
+oid sha256:60e666ef654e5df1af719d63e0bb63b506e3c58ebbe60473d3b20228b41df7a6
 size 163832792

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16d4f44bc57c80b7f44a5dca2132f7a5cb26b5afa80f8b76663245880cc40f00
+oid sha256:4a6d75383dd0ddba0bcfbc60b4eb17267bfdbe16b571d820fda5cec5e8c9fb6b
 size 5560