File size: 1,481 Bytes
840b176
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
# Shakespeare GPT

A GPT-2 model fine-tuned on Shakespeare's works, capable of generating Shakespeare-style text.

## Project Overview

This project implements a GPT-2 architecture trained on Shakespeare's works to generate Shakespeare-style text. The model uses a context window of 1024 tokens and implements various optimizations including gradient accumulation and learning rate scheduling.

## Model Architecture

- Base Architecture: GPT-2 (124M parameters)
- Layers: 12
- Heads: 12
- Embedding Dimension: 768
- Context Length: 1024 tokens
- Total Parameters: ~124M

## Training Details

- Dataset: Shakespeare's complete works
- Training Device: GPU/MPS (Apple Silicon)
- Batch Size: 16 (Effective batch size: 64 with gradient accumulation)
- Learning Rate: 6e-4 with cosine decay
- Weight Decay: 0.1
- Training Steps: 10,000

## Performance

- Best Validation Loss: [Insert your best validation loss]
- Training Time: [Insert your training time]

## Requirements
- bash
- pip install -r requirements.txt

## Project Structure
β”œβ”€β”€ src/
β”‚ β”œβ”€β”€ train_shakespeare.py # Training script

β”‚ β”œβ”€β”€ app.py # Gradio interface

β”‚ └── input.txt # Training data

β”œβ”€β”€ requirements.txt

└── README.md



## Usage



### Training



To train the model:



bash

python src/train_shakespeare.py


### Inference

- To run the Gradio interface locally:
- bash
- python src/app.py

bash
python src/app.py