End of training

README.md

@@ -1,7 +1,9 @@
 ---
-library_name: transformers
-license: apache-2.0
 base_model: distilbert/distilgpt2
+datasets:
+- wikimedia/wikipedia
+library_name: Distily
+license: apache-2.0
 tags:
 - generated_from_trainer
 model-index:
@@ -9,50 +11,126 @@ model-index:
   results: []
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
 
-# distily_validate_extra_grad_stats
+# Summary
 
-This model is a fine-tuned version of [distilbert/distilgpt2](https://huggingface.co/distilbert/distilgpt2) on an unknown dataset.
-It achieves the following results on the evaluation set:
-- Loss: 1.7150
+Distilled with [Distily](https://github.com/lapp0/distily) library
+using teacher model [gpt2](https://huggingface.co/gpt2)
+on dataset [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia).
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment.
 
-## Model description
+# Model description
 
 More information needed
 
-## Intended uses & limitations
+# Intended uses & limitations
 
 More information needed
+-->
 
-## Training and evaluation data
+# Model Architecture:
+- **Architecture**: `GPT2LMHeadModel`
+- **Total Parameters**: 81,912,576
+- **Data Type (dtype)**: torch.bfloat16
+- **Model Size**: 0.16 GB
 
-More information needed
 
-## Training procedure
+# Benchmark Metrics Comparison
 
-### Training hyperparameters
+| Metric |  |
+| :--- |
 
-The following hyperparameters were used during training:
-- 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
-- num_epochs: 1.0
+# Resource Usage Comparison
+
+- VRAM Use: 7.4259 GB
+
+# Distillation (Teacher -> Student) Architecture Difference:
 
-### Training results
+- **Architecture**: `GPT2LMHeadModel` -> `GPT2LMHeadModel`
+- **Total Parameters**: 124,439,808 -> 81,912,576
+- **Data Type (dtype)**: torch.bfloat16 -> torch.bfloat16
+- **Model Size**: 0.24 GB -> 0.16 GB
 
-| Training Loss | Epoch | Step | Validation Loss |
-|:-------------:|:-----:|:----:|:---------------:|
-| No log        | 0     | 0    | 21.3700         |
+<details>
+<summary>Module Diff Details</summary>
 
+```diff
+--- teacher model modules
++++ student model modules
+@@ -4,7 +4,7 @@
+     (wpe): Embedding(1024, 768)
+     (drop): Dropout(p=0.1, inplace=False)
+     (h): ModuleList(
+-      (0-11): 12 x GPT2Block(
++      (0-5): 6 x GPT2Block(
+         (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
+         (attn): GPT2FlashAttention2(
+           (c_attn): Conv1D()
 
-### Framework versions
+```
+
+</details>
+<br/>
+
+# Train Dataset
+Trained on 6,813,447 tokens from the [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dataset.
+
+- Num Samples: `9,900`
+- Subset: `20231101.en`
+- Split: `train`
+
+
+# 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))
+```
+
+# Hyperparameters
+The following hyperparameters were used during training:
 
+<details>
+<summary>Expand</summary>
+
+- 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`
+- 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))`
+- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7feee8b1d330>`
+- student_model_name_or_path: `None`
+- student_config_name_or_path: `distilbert/distilgpt2`
+- student_model_config: `None`
+- reinitialize_weights: `None`
+- copy_teacher_modules: `[('lm_head', False)]`
+- student_model_as_bitnet: `False`
+- teacher_model_name_or_path: `gpt2`
+- teacher_load_in_8bit: `False`
+- teacher_load_in_4bit: `False`
+- dataset_uri: `wikimedia/wikipedia`
+- dataset_subset: `20231101.en`
+- dataset_split: `train`
+- dataset_column_name: `text`
+- dataset_sample_size: `10000`
+- dataset_test_size: `0.01`
+- gradient_accumulation_steps: `1`
+- weight_decay: `0.0`
+- max_grad_norm: `1.0`
+- warmup_ratio: `0`
+- warmup_steps: `0`
+- gradient_checkpointing: `True`
+
+</details>
+<br/>
+
+
+# Framework Versions
+- Distily 0.4.1
 - Transformers 4.44.2
 - Pytorch 2.3.0
 - Datasets 2.21.0
-- Tokenizers 0.19.1

logs/attn_projector=orthogonal, attn_weight=5, extra_grad_stats=True, learning_rate=0.0002, per_device_train_batch_size=4, warmup_ratio=0/events.out.tfevents.1725293781.261a4d6fb516

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:147595160dca32431da2e8b342abb2fe84f01be7c8fd6ccdeb2c71bf1ad57cde
+size 249