app.py

import gradio as gr
import torch
from transformers import GPT2Tokenizer

# Define your model
class GPTLanguageModel(torch.nn.Module):
    def __init__(self, vocab_size, hidden_size):
        super(GPTLanguageModel, self).__init__()
        self.embedding = torch.nn.Embedding(vocab_size, hidden_size)
        # Add other layers as needed

    def forward(self, input_ids):
        return self.embedding(input_ids)  # Placeholder for the forward pass

    def generate(self, input_ids, max_length=100):
        # Custom generation logic here
        return input_ids  # Placeholder

# Define the Custom Text Generation Pipeline
class CustomTextGenerationPipeline:
    def __init__(self, model, tokenizer):
        self.model = model
        self.tokenizer = tokenizer

    def __call__(self, prompt, max_length=100):
        input_ids = self.tokenizer.encode(prompt, return_tensors='pt')
        generated_ids = self.model.generate(input_ids, max_length=max_length)
        return self.tokenizer.decode(generated_ids[0], skip_special_tokens=True)

# Load tokenizer and model
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")  # Or your custom tokenizer
vocab_size = tokenizer.vocab_size
model = GPTLanguageModel(vocab_size=vocab_size, hidden_size=768)  # Set sizes appropriately

# Load model weights
try:
    model.load_state_dict(torch.load("model.pth", map_location=torch.device('cpu'), weights_only=True), strict=False)
except RuntimeError as e:
    print(f"Error loading model weights: {e}")

model.eval()

# Create the pipeline
pipeline = CustomTextGenerationPipeline(model, tokenizer)

# Define the Gradio response function
def respond(message):
    return pipeline(message, max_length=100)

# Create the Gradio interface
demo = gr.Interface(
    fn=respond,
    inputs=gr.Textbox(lines=2, placeholder="Enter your prompt..."),
    outputs="text",
)

if __name__ == "__main__":
    demo.launch()