---
base_model: gpt2
datasets:
- wikimedia/wikipedia
library_name: Distily
license: mit
tags:
- bitnet
- 1.58b
- generated_from_trainer
model-index:
- name: distily_multi_experiment
  results: []
---


# Summary

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

More information needed

# Intended uses & limitations

More information needed
-->

# Model Architecture:
- **Architecture**: `GPT2LMHeadModel`
- **Total Parameters**: 124,439,808
- **Data Type (dtype)**: torch.bfloat16
- **Model Size**: 0.24 GB


# Evaluation Metrics Comparison

| step | epoch | enwikippl | frwikippl | loss | runtime | samples_per_second | steps_per_second | tinystoriesppl | zhwikippl |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| **teacher eval** |  | 43.25 | 61.25 |  |  |  |  | 11.6875 | 19.125 |
| 0 | 0 | 2473901162496.0 | 170424302305280.0 | 45.7764 | 30.2746 | 82.577 | 10.339 | 4060086272.0 | 71468255805440.0 |
| 2500 | 0.0404 | 2040.0 | 20608.0 | 20.8472 | 30.2064 | 82.764 | 10.362 | 1472.0 | 60160.0 |
| 5000 | 0.0808 | 488.0 | 3120.0 | 18.4130 | 30.6595 | 81.541 | 10.209 | 338.0 | 1112.0 |
| 7500 | 0.1212 | 276.0 | 1296.0 | 17.0012 | 30.289 | 82.538 | 10.334 | 255.0 | 249.0 |
| 10000 | 0.1616 | 202.0 | 756.0 | 16.2078 | 30.2709 | 82.588 | 10.34 | 188.0 | 304.0 |
| 12500 | 0.2020 | 145.0 | 540.0 | 15.0996 | 30.2185 | 82.731 | 10.358 | 131.0 | 176.0 |
| 15000 | 0.2424 | 123.5 | 490.0 | 14.5283 | 30.2533 | 82.636 | 10.346 | 93.0 | 146.0 |
| 17500 | 0.2828 | 95.0 | 376.0 | 14.1520 | 30.2403 | 82.671 | 10.35 | 75.5 | 137.0 |
| 20000 | 0.3232 | 79.0 | 306.0 | 13.6446 | 30.1204 | 83.0 | 10.392 | 63.25 | 130.0 |
| 22500 | 0.3636 | 66.0 | 219.0 | 13.1452 | 30.158 | 82.897 | 10.379 | 50.0 | 80.5 |
| 25000 | 0.4040 | 63.0 | 200.0 | 12.9619 | 30.1269 | 82.982 | 10.389 | 43.75 | 77.5 |
| 27500 | 0.4444 | 59.0 | 197.0 | 12.8388 | 30.3214 | 82.45 | 10.323 | 40.5 | 73.5 |
| 30000 | 0.4848 | 59.5 | 204.0 | 12.8191 | 30.3164 | 82.464 | 10.324 | 40.5 | 70.5 |
| 32500 | 0.5253 | 58.25 | 176.0 | 12.7778 | 30.2231 | 82.718 | 10.356 | 38.75 | 61.75 |
| 35000 | 0.5657 | 58.25 | 169.0 | 12.6562 | 30.35 | 82.372 | 10.313 | 36.5 | 45.5 |
| 37500 | 0.6061 | 56.75 | 158.0 | 12.6014 | 30.3685 | 82.322 | 10.307 | 37.0 | 50.5 |
| 40000 | 0.6465 | 55.0 | 156.0 | 12.5674 | 30.3598 | 82.346 | 10.31 | 33.75 | 59.5 |
| 42500 | 0.6869 | 54.5 | 147.0 | 12.5141 | 30.3209 | 82.451 | 10.323 | 34.25 | 52.5 |
| 45000 | 0.7273 | 50.75 | 135.0 | 12.2860 | 30.244 | 82.661 | 10.349 | 29.5 | 41.75 |
| 47500 | 0.7677 | 50.5 | 127.0 | 12.2408 | 30.3366 | 82.409 | 10.318 | 28.875 | 35.0 |
| 50000 | 0.8081 | 50.25 | 125.5 | 12.2160 | 30.2563 | 82.627 | 10.345 | 28.625 | 39.0 |
| 52500 | 0.8485 | 49.25 | 123.0 | 12.1936 | 30.2253 | 82.712 | 10.356 | 28.5 | 35.5 |
| 55000 | 0.8889 | 49.25 | 121.0 | 12.1620 | 30.1898 | 82.81 | 10.368 | 27.875 | 35.0 |
| 57500 | 0.9293 | 48.75 | 120.0 | 12.1488 | 30.2559 | 82.628 | 10.345 | 27.75 | 33.5 |
| 60000 | 0.9697 | 48.75 | 119.5 | 12.1404 | 30.1517 | 82.914 | 10.381 | 27.625 | 33.25 |
| 61875 | 1.0 | 48.75 | 120.0 | 12.1402 | 30.2129 | 82.746 | 10.36 | 27.625 | 33.5 |

# Resource Usage Comparison

- VRAM Use: 7.7831 GB

`# Distillation (Teacher -> Student) Architecture Difference:

- **Architecture**: `GPT2LMHeadModel` -> `GPT2LMHeadModel`
- **Total Parameters**: 124,439,808 -> 124,439,808
- **Data Type (dtype)**: 124439808 -> torch.bfloat16
- **Model Size**: 0.24 GB -> 0.24 GB

<details>
<summary>Module Diff Details</summary>

```diff

```

</details>
<br/>

# Train Dataset
Trained on 145,744,973 tokens from the [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dataset.

- Num Samples: `247,500`
- 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=25.0, loss_fn=cos, layer_mapper=layer-2))
```

# Hyperparameters
The following hyperparameters were used during training:

<details>
<summary>Expand</summary>

- learning_rate: `0.0001`
- 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: `linear`
- 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=cos, layer_mapper=layer-2))`
- train_embeddings: `True`
- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7fbd2823f2b0>`
- student_model_name_or_path: `None`
- student_config_name_or_path: `None`
- student_model_config: `None`
- reinitialize_weights: `None`
- copy_teacher_modules: `[('lm_head', False)]`
- student_model_as_bitnet: `True`
- student_model_compile: `False`
- dropout: `None`
- teacher_model_name_or_path: `gpt2`
- teacher_load_in_8bit: `False`
- teacher_load_in_4bit: `False`
- teacher_model_compile: `False`
- dataset_uri: `wikimedia/wikipedia`
- dataset_subset: `20231101.en`
- dataset_split: `train`
- dataset_column_name: `text`
- dataset_sample_size: `250000`
- dataset_test_size: `0.01`
- gradient_accumulation_steps: `1`
- weight_decay: `0.0`
- max_grad_norm: `1.0`
- warmup_ratio: `0.5`
- warmup_steps: `0`
- gradient_checkpointing: `True`

</details>
<br/>


# Framework Versions
- Distily 0.2.0
- Transformers 4.44.0
- Pytorch 2.3.0
- Datasets 2.21.0