Create app.py

app.py

@@ -0,0 +1,96 @@
+import gradio as gr
+from transformers import pipeline
+import gc
+import os
+import shutil
+import torch
+import psutil
+
+# Define path for HF cache to clean
+HF_CACHE_DIR = os.path.expanduser("~/.cache/huggingface/hub")
+
+def get_system_stats():
+    """Returns a dictionary of current system metrics with formatted strings."""
+    mem = psutil.virtual_memory()
+    disk = psutil.disk_usage('/')
+    return {
+        "CPU": f"{psutil.cpu_percent(interval=1)}%",
+        "Memory": f"{round(mem.used / (1024**3), 2)} / {round(mem.total / (1024**3), 2)} GB",
+        "Disk": f"{round(disk.used / (1024**3), 2)} / {round(disk.total / (1024**3), 2)} GB"
+    }
+
+def load_new_model(model_id):
+    # Clear old model from memory
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+    
+    try:
+        # Load a text-generation pipeline
+        pipe = pipeline("text-generation", model=model_id)
+        return pipe, f"Successfully loaded {model_id}"
+    except Exception as e:
+        return None, f"Error loading model: {str(e)}"
+
+def run_inference(model, system_prompt, user_prompt):
+    if not model:
+        return "Please load a model first."
+    
+    # Combine prompts
+    full_prompt = f"System: {system_prompt}\nUser: {user_prompt}"
+    
+    # Run inference
+    result = model(full_prompt, max_new_tokens=50)
+    return result[0]['generated_text']
+
+def clean_cache():
+    if os.path.exists(HF_CACHE_DIR):
+        shutil.rmtree(HF_CACHE_DIR)
+        os.makedirs(HF_CACHE_DIR)
+        return "Cache cleaned successfully!"
+    return "Cache directory not found."
+
+# Gradio Interface
+with gr.Blocks(title="Small Model Tester") as app:
+    gr.Markdown("# Advanced HF Model Tester (Free Tier)")
+    
+    # Stats Section
+    with gr.Accordion("System Monitoring", open=True):
+        stats_output = gr.JSON(label="Live System Stats")
+        # Gradio timer to update stats every 5 seconds
+        gr.Timer(5).tick(get_system_stats, None, stats_output)
+    
+    current_model = gr.State(None)
+    
+    with gr.Row():
+        model_id_input = gr.Textbox(label="Model ID", value="distilgpt2", placeholder="e.g., distilgpt2, tiny-llama/tiny-llama-1.1b")
+        load_btn = gr.Button("Load Model")
+    
+    status_output = gr.Markdown("Status: Waiting to load model...")
+    
+    with gr.Row():
+        system_prompt = gr.Textbox(label="System Prompt", placeholder="You are a helpful assistant.")
+        user_prompt = gr.Textbox(label="User Prompt", placeholder="Hello!")
+    
+    run_btn = gr.Button("Run Inference", variant="primary")
+    output_text = gr.Textbox(label="Result")
+    
+    clean_btn = gr.Button("Clean Cache")
+    
+    # Events
+    load_btn.click(
+        fn=load_new_model, 
+        inputs=[model_id_input], 
+        outputs=[current_model, status_output]
+    )
+    
+    run_btn.click(
+        fn=run_inference,
+        inputs=[current_model, system_prompt, user_prompt],
+        outputs=[output_text]
+    )
+    
+    clean_btn.click(fn=clean_cache, outputs=[status_output])
+
+if __name__ == "__main__":
+    app.launch()