GPT-2 SFT on OASST1

This model is a fine-tuned version of openai-community/gpt2 on the OpenAssistant/oasst1 dataset.

Training Procedure

• Base Model: GPT-2 Small (124M parameters)
• Dataset: OpenAssistant OASST1 (English only)
• Format: user: ... assistant: <s> ... </s>
• Epochs: 15
• Learning Rate: 5e-5
• Batch Size: 4 (effective 16 with gradient accumulation)

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "gitcoreai/gpt-2-small-sft"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

prompt = "user: hello, how are you?\nassistant: <s>"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=60)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))