distily/distily_attn_mlp_sweep

Summary

Distilled with Distily library using teacher model gpt2 on dataset wikimedia/wikipedia.

Model Architecture:

• Architecture: GPT2LMHeadModel
• Total Parameters: 81,912,576
• Data Type (dtype): torch.bfloat16
• Model Size: 0.16 GB

Student Model Details

GPT2LMHeadModel(
  (transformer): GPT2Model(
    (wte): Embedding(50257, 768)
    (wpe): Embedding(1024, 768)
    (drop): Dropout(p=0.1, inplace=False)
    (h): ModuleList(
      (0-5): 6 x GPT2Block(
        (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        (attn): GPT2FlashAttention2(
          (c_attn): Conv1D()
          (c_proj): Conv1D()
          (attn_dropout): Dropout(p=0.1, inplace=False)
          (resid_dropout): Dropout(p=0.1, inplace=False)
        )
        (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        (mlp): GPT2MLP(
          (c_fc): Conv1D()
          (c_proj): Conv1D()
          (act): NewGELUActivation()
          (dropout): Dropout(p=0.1, inplace=False)
        )
      )
    )
    (ln_f): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
  )
  (lm_head): Linear(in_features=768, out_features=50257, bias=False)
)

Resource Usage

• Max Train VRAM Use: 15.2844 GB
• Available VRAM: 23.6429 GB
• GPUs:
  • 1x NVIDIA GeForce RTX 4090
• CPUs: 128
• CPU Memory: 503.5412 GB
• CPU Memory Bandwidth: 3200 GB/s

Distillation (Teacher -> Student) Architecture Difference:

• 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

Module Diff Details

--- 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()

Train Dataset

Trained on 521,334,462 tokens from the wikimedia/wikipedia dataset.

• Num Samples: 990,000
• Subset: 20231101.en
• Split: train

Training Objective

DistillationObjective(
    logits_loss_component=LossComponent(
        weight=1,
        loss_fn='kl'
    ),
    hs_loss_component=LossComponent(
        weight=0
    ),
    attn_loss_component=LossComponent(
        weight=0
    )
)

Hyperparameters

The following hyperparameters were used during training:

Expand

• learning_rate: 0.0002
• train_batch_size: 16
• 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( weight=1, loss_fn='kl' ), hs_loss_component=LossComponent( weight=0 ), attn_loss_component=LossComponent( weight=0 ) )
• lr_scheduler: <torch.optim.lr_scheduler.LambdaLR object at 0x7facb65ee170>
• 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: 1000000
• dataset_test_size: 0.01
• gradient_accumulation_steps: 1
• weight_decay: 0.0
• max_grad_norm: 1.0
• warmup_ratio: 0.0
• warmup_steps: 0
• gradient_checkpointing: True

Framework Versions

• Distily 0.5.0
• Transformers 4.44.2
• Pytorch 2.5.0.dev20240905+cu121
• Datasets 2.18.0