Mini GPT1 Clone

This is a decoder-only transformer model (GPT1-style) trained from scratch using PyTorch.

Model Details

Architecture: Decoder-only Transformer
Layers: 6
Embedding Size: 512
Heads: 8
Feedforward Dim: 2048
Sequence Length: 256
Vocab Size: 35,000

Tokenizer

Trained using ByteLevelBPETokenizer from the tokenizers library.

Inference Example

from transformers import PreTrainedTokenizerFast, AutoModelForCausalLM
import torch

tokenizer = PreTrainedTokenizerFast(tokenizer_file="tokenizer/tokenizer.json")
model = AutoModelForCausalLM.from_pretrained("dilip025/mini-gpt1")

prompt = "Once upon a time,"
input_ids = tokenizer(prompt, return_tensors="pt").input_ids
outputs = model.generate(input_ids, max_length=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
License
MIT