---
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 | 30.1483 | 82.923 | 10.382 | 4060086272.0 | 71468255805440.0 |
| 2500 | 0.0404 | 2192.0 | 21888.0 | 20.8264 | 30.1218 | 82.996 | 10.391 | 1280.0 | 91136.0 |
| 5000 | 0.0808 | 486.0 | 3120.0 | 18.4124 | 30.4931 | 81.986 | 10.265 | 338.0 | 1128.0 |
| 7500 | 0.1212 | 276.0 | 1296.0 | 17.0012 | 30.1359 | 82.958 | 10.386 | 255.0 | 250.0 |
| 10000 | 0.1616 | 202.0 | 756.0 | 16.2078 | 30.1737 | 82.854 | 10.373 | 189.0 | 310.0 |
| 12500 | 0.2020 | 145.0 | 540.0 | 15.1002 | 30.1534 | 82.909 | 10.38 | 130.0 | 177.0 |
| 15000 | 0.2424 | 123.5 | 488.0 | 14.5297 | 30.1065 | 83.039 | 10.396 | 93.0 | 149.0 |
| 17500 | 0.2828 | 95.0 | 376.0 | 14.1518 | 30.1635 | 82.882 | 10.377 | 75.5 | 137.0 |
| 20000 | 0.3232 | 79.0 | 308.0 | 13.6434 | 30.1294 | 82.976 | 10.389 | 63.0 | 147.0 |
| 22500 | 0.3636 | 67.0 | 217.0 | 13.1501 | 30.1133 | 83.02 | 10.394 | 51.0 | 83.5 |
| 25000 | 0.4040 | 63.75 | 203.0 | 12.9684 | 30.2115 | 82.75 | 10.36 | 43.75 | 69.5 |
| 27500 | 0.4444 | 59.0 | 195.0 | 12.8408 | 30.1817 | 82.832 | 10.371 | 40.0 | 76.0 |
| 30000 | 0.4848 | 59.0 | 194.0 | 12.8237 | 30.1805 | 82.835 | 10.371 | 41.5 | 55.75 |
| 32500 | 0.5253 | 58.5 | 174.0 | 12.7748 | 30.1793 | 82.838 | 10.371 | 38.25 | 61.25 |
| 35000 | 0.5657 | 57.25 | 171.0 | 12.6602 | 30.1387 | 82.95 | 10.385 | 36.25 | 43.75 |
| 37500 | 0.6061 | 56.25 | 159.0 | 12.5996 | 30.2154 | 82.739 | 10.359 | 36.75 | 49.5 |
| 40000 | 0.6465 | 55.0 | 156.0 | 12.5526 | 30.1918 | 82.804 | 10.367 | 33.75 | 58.5 |
| 42500 | 0.6869 | 54.5 | 148.0 | 12.4990 | 30.1749 | 82.85 | 10.373 | 34.0 | 48.25 |
| 45000 | 0.7273 | 50.75 | 134.0 | 12.2851 | 30.1611 | 82.888 | 10.378 | 29.0 | 42.0 |
| 47500 | 0.7677 | 51.25 | 126.5 | 12.2410 | 30.2021 | 82.776 | 10.364 | 28.75 | 36.0 |
| 50000 | 0.8081 | 50.5 | 124.0 | 12.2169 | 30.166 | 82.875 | 10.376 | 28.5 | 38.25 |
| 52500 | 0.8485 | 49.25 | 122.0 | 12.1910 | 30.1439 | 82.936 | 10.384 | 28.25 | 34.5 |
| 55000 | 0.8889 | 49.25 | 120.5 | 12.1604 | 30.1149 | 83.015 | 10.394 | 27.625 | 35.0 |
| 57500 | 0.9293 | 49.25 | 120.5 | 12.1473 | 30.1142 | 83.017 | 10.394 | 27.75 | 33.75 |
| 60000 | 0.9697 | 49.0 | 119.5 | 12.1403 | 30.143 | 82.938 | 10.384 | 27.625 | 33.5 |
| 61875 | 1.0 | 49.0 | 120.0 | 12.1391 | 30.2511 | 82.642 | 10.347 | 27.625 | 33.75 |

# 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 0x7fbeac4d9c00>`
- 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