Himanshu13x/gpt2-medium355M-sft

GPT-2 (From Scratch in PyTorch) — Fine-Tuned Version

This model is a custom GPT-2 implementation built entirely from scratch in PyTorch (no Hugging Face Transformers for the architecture itself) and fine-tuned on a custom dataset using Supervised Fine-Tuning (SFT).

Model Details

• Architecture: GPT-2 (from scratch)
• Variants Supported: gpt2-small, gpt2-medium, gpt2-large, gpt2-xl
• Framework: PyTorch
• Pretraining Source: Loaded GPT-2 pretrained weights from OpenAI format
• Fine-Tuning Method: Supervised Fine-Tuning (SFT)
• Fine-Tuning Data: Custom dataset (domain-specific; see dataset section)
• Tokenization: GPT-2 tokenizer style

Repo

• https://github.com/Himanshu-00/gpt2-from-scratch-pytorch.git