from transformers import GPT2LMHeadModel, GPT2TokenizerFast, pipeline
import gradio as gr
import os

# Set the correct path to your model directory on Hugging Face Hub
model_dir = "JakeTurner616/Adonalsium-gpt2"

# Manually specify the model's configuration and weights files
model = GPT2LMHeadModel.from_pretrained(model_dir, torch_dtype='auto', low_cpu_mem_usage=True)
tokenizer = GPT2TokenizerFast.from_pretrained(model_dir, use_fast=True)

# Ensure the tokenizer uses the correct pad token
tokenizer.pad_token = tokenizer.eos_token

# Create the text generation pipeline
generator = pipeline('text-generation', model=model, tokenizer=tokenizer)

def generate_text(prompt):
    # Generate text
    generated_texts = generator(prompt, max_length=150, temperature=0.7,
                                top_p=0.85, repetition_penalty=1.3,
                                num_return_sequences=1, no_repeat_ngram_size=2,
                                early_stopping=True)
    return generated_texts[0]['generated_text']

# Create the Gradio interface
iface = gr.Interface(fn=generate_text, inputs="text", outputs="text", 
                     title="Cosmere Text Generator", 
                     description="Generate text based on the Cosmere series by Brandon Sanderson.")
# Launch the interface
iface.launch()