Training in progress, step 99000

README.md

@@ -44,7 +44,7 @@ More information needed
 
 # Resource Usage Comparison
 
-- VRAM Use: 8.0722 GB
+- VRAM Use: 7.4167 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, norm=instancenorm, projector=orthogonal))
+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, norm=rmsnorm, projector=mlp))
 ```
 
 # Hyperparameters
@@ -95,15 +95,15 @@ The following hyperparameters were used during training:
 <summary>Expand</summary>
 
 - learning_rate: `0.0002`
-- train_batch_size: `8`
+- 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`
 - 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, norm=instancenorm, projector=orthogonal))`
+- 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, norm=rmsnorm, projector=mlp))`
 - train_embeddings: `True`
-- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7f6b219afd30>`
+- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7f350be93280>`
 - student_model_name_or_path: `None`
 - student_config_name_or_path: `distilbert/distilgpt2`
 - student_model_config: `None`
@@ -134,5 +134,5 @@ The following hyperparameters were used during training:
 # Framework Versions
 - Distily 0.4.1
 - Transformers 4.44.2
-- Pytorch 2.3.0
-- Datasets 2.21.0
+- Pytorch 2.4.0+cu121
+- Datasets 2.18.0

logs/attn_norm=rmsnorm_teacher_only, attn_projector=mlp, attn_weight=5, learning_rate=0.0002, per_device_train_batch_size=4, warmup_ratio=0/events.out.tfevents.1725166327.a7e428977e35

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4336ffd7133e98ab95c89fbcc26cf904c878f05549e3825db1ded537b7cb9544
+size 47486062

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37928efaf6e1955069debf9065abe3795dbaaf09ac1735d23f28a942a3008baf
+oid sha256:46a872a19810cb4dcebd049e7f2be8481554c1efb9f12e7adbe3e9c6ff3663b6
 size 163832792

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6d2c2667d6e25d0055e7408133a5d519bff47baef4365bdc3c0e5227047da28
+oid sha256:d1baa9bff7f36cd037e89bcbc4f92dc4ed7a5dd5d116a06e1eabe19580114ef9
 size 5560