Upload 2 files

app.py

@@ -1,34 +1,29 @@
-import os
 import torch
-import gradio as gr
-from gpt_dev import GPTLanguageModel, encode, decode, stoi, itos
+import os
+from gpt_dev import GPTLanguageModel, encode, decode, generate_text  # Import necessary parts from gpt_dev.py
 
-# Load the model
+# Set up device (GPU or CPU)
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-model = GPTLanguageModel().to(device)
 
-# Load the trained model state
+# Load your saved model (pre-trained model's state dict)
+block_size= 256
+model_type = "gpt"
+n_embd = 384
+n_head = 6
+n_layer = 6
+transformers_version = 4.44.2
+vocab_size = 95
+
+# Instantiate the model
+model = GPTLanguageModel(vocab_size, embedding_size, num_heads, num_layers)
+model.to(device)
+
+# Load the pre-trained weights
 if os.path.exists("gpt_language_model.pth"):
     checkpoint = torch.load("gpt_language_model.pth", map_location=device)
     model.load_state_dict(checkpoint)
 
-model.eval()  # Set model to evaluation mode
-
-def get_response(prompt):
-    # Encode the prompt
-    encoded_prompt = encode(prompt)
-    input_tensor = torch.tensor(encoded_prompt).unsqueeze(0).to(device)  # Shape (1, seq_length)
-
-    # Generate response
-    with torch.no_grad():
-        output = model(input_tensor)  # Generate the output
-        output_text = decode(output[0].tolist())  # Decode the output to text
-
-    return output_text
-
-def main():
-    iface = gr.Interface(fn=get_response, inputs="text", outputs="text", title="GPT Language Model")
-    iface.launch()
-
-if __name__ == "__main__":
-    main()
+# Generate text based on a prompt
+start_prompt = "Once upon a time"
+generated_text = generate_text(model, start_prompt, max_length=100)
+print(generated_text)  # Display the generated text

gpt_dev.py

@@ -426,40 +426,19 @@ b = torch.randint(0, 10, (3, 2)).float()
 c = a @ b
 
 def generate_text(model, start_prompt, max_length=100, temperature=1.0):
-    # Encode the input prompt into tensor format
     input_ids = torch.tensor(encode(start_prompt), dtype=torch.long).unsqueeze(0).to(device)
-
-    # Set model to evaluation mode
     model.eval()
-
-    # Generated tokens list (starting with the prompt)
     generated_ids = input_ids.tolist()[0]
-
-    # Sampling loop
     with torch.no_grad():
         for _ in range(max_length):
-            # Get model predictions for the current input
             logits, _ = model(input_ids)
-
-            # Optionally, divide logits by temperature to control randomness
             logits = logits[:, -1, :] / temperature
-
-            # Apply softmax to get probabilities and sample the next token
             probs = torch.nn.functional.softmax(logits, dim=-1)
             next_token = torch.multinomial(probs, num_samples=1)
-
-            # Append the generated token to the list
             generated_ids.append(next_token.item())
-
-            # Update the input with the new token for the next iteration
             input_ids = torch.cat((input_ids, next_token), dim=1)
-
-    # Decode the generated tokens back into text
-    generated_text = decode(generated_ids)
-    return generated_text
-
-# Example usage
-start_prompt = "Once upon a time"
-generated_text = generate_text(m, start_prompt, max_length=100)
+    return decode(generated_ids)
 
 
+if __name__ == "__main__":
+    train_model()