asdasdasdasd

app.py

@@ -1,62 +1,94 @@
+import json
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
-from huggingface_hub import HfApi
-import gradio as gr
-import os
 from peft import PeftModel, PeftConfig
+from huggingface_hub import HfApi, Repository
+import gradio as gr
+
+# Global variables to store the model and tokenizer
+merged_model = None
+tokenizer = None
 
-# Function to merge models using PEFT
 def merge_models():
-    base_model_name = "meta-llama/Meta-Llama-3-8B"
-    finetuned_model_name = "NoaiGPT/autotrain-14mrs-fc44l"
+    print("Loading base model...")
+    base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
     
-    # Load the base model
-    base_model = AutoModelForCausalLM.from_pretrained(base_model_name)
-    base_tokenizer = AutoTokenizer.from_pretrained(base_model_name)
+    print("Loading fine-tuned model...")
+    peft_config = PeftConfig.from_pretrained("NoaiGPT/autotrain-14mrs-fc44l")
+    fine_tuned_model = PeftModel.from_pretrained(base_model, peft_config)
     
-    # Load the fine-tuned model using PEFT
-    peft_config = PeftConfig.from_pretrained(finetuned_model_name)
-    finetuned_model = PeftModel.from_pretrained(base_model, peft_config)
+    print("Merging models...")
+    merged_model = fine_tuned_model.merge_and_unload()
     
-    # Merge the models (simple weight averaging here for demonstration; adjust as needed)
-    for param_base, param_finetuned in zip(base_model.parameters(), finetuned_model.parameters()):
-        param_base.data = (param_base.data + param_finetuned.data) / 2
+    print("Saving merged model...")
+    merged_model.save_pretrained("merged_model")
     
-    # Save the merged model
-    merged_model_name = "./merged_model"
-    base_model.save_pretrained(merged_model_name)
-    base_tokenizer.save_pretrained(merged_model_name)
+    print("Saving tokenizer...")
+    tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
+    tokenizer.save_pretrained("merged_model")
     
-    return merged_model_name
+    return merged_model, tokenizer
 
-# Function to upload the merged model to Hugging Face Hub
-def upload_to_hf(repo_id, merged_model_name):
+def push_to_hub(repo_name):
+    print(f"Pushing merged model to Hugging Face Hub: {repo_name}")
     api = HfApi()
-    model_files = [os.path.join(merged_model_name, f) for f in os.listdir(merged_model_name)]
-    
-    for file in model_files:
-        api.upload_file(
-            path_or_fileobj=file,
-            path_in_repo=os.path.basename(file),
-            repo_id=repo_id,
-            repo_type="model"
-        )
-
-    return f"Model uploaded to Hugging Face Hub at {repo_id}."
-
-# Gradio function to handle the merge button click
-def merge_button_clicked():
-    merged_model_name = merge_models()
-    repo_id = "NoaiGPT/autotrain-14mrs-fc44l"
-    return upload_to_hf(repo_id, merged_model_name)
-
-# Create the Gradio interface
-with gr.Blocks() as demo:
-    with gr.Row():
-        merge_button = gr.Button("Merge Models")
-        output = gr.Textbox(label="Output")
-
-    merge_button.click(merge_button_clicked, outputs=output)
-
-# Launch the Gradio interface
-demo.launch()
+    
+    try:
+        api.create_repo(repo_name, private=True)
+        print(f"Created new repository: {repo_name}")
+    except Exception as e:
+        print(f"Repository already exists or error occurred: {e}")
+    
+    api.upload_folder(
+        folder_path="merged_model",
+        repo_id=repo_name,
+        repo_type="model",
+    )
+    print("Model pushed successfully!")
+
+def generate_text(input_text):
+    global merged_model, tokenizer
+    
+    if merged_model is None or tokenizer is None:
+        return "Model not loaded. Please run the merge process first."
+    
+    input_ids = tokenizer.encode(input_text, return_tensors="pt")
+    
+    with torch.no_grad():
+        output = merged_model.generate(input_ids, max_length=200, num_return_sequences=1)
+    
+    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
+    return generated_text
+
+def run_merge_and_push():
+    global merged_model, tokenizer
+    
+    # Merge models
+    merged_model, tokenizer = merge_models()
+    
+    # Push to Hugging Face Hub
+    hub_repo_name = "your-username/merged-llama3-8b-instruct"
+    push_to_hub(hub_repo_name)
+    
+    return "Model merged and pushed to Hugging Face Hub successfully!"
+
+# Gradio interface
+iface = gr.Interface(
+    fn=generate_text,
+    inputs=gr.Textbox(lines=5, label="Input Text"),
+    outputs=gr.Textbox(lines=10, label="Generated Text"),
+    title="Merged Llama 3 8B Instruct Model",
+    description="Enter text to generate a response from the merged model.",
+)
+
+merge_and_push_button = gr.Button("Merge Models and Push to Hub")
+merge_and_push_output = gr.Textbox(lines=20, label="Merge and Push Output")
+
+merge_and_push_button.click(run_merge_and_push, outputs=merge_and_push_output)
+
+# Launch the interface
+if __name__ == "__main__":
+    gr.TabbedInterface(
+        [iface, gr.Interface(fn=lambda: "", inputs=None, outputs=merge_and_push_output)],
+        ["Generate Text", "Merge and Push"]
+    ).launch()