Leetmonkey In Action. Darn LeetMonkey these days

app.py

@@ -4,34 +4,51 @@ from llama_cpp import Llama
 import re
 from datasets import load_dataset
 import random
+import logging
+import os
 import autopep8
-import textwrap
-import threading
+
+# Set up logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
 
 # Define the model options
 gguf_models = {
+    "Q8_0": "leetmonkey_peft__q8_0.gguf",
     "Exact Copy": "leetmonkey_peft_exact_copy.gguf",
     "F16": "leetmonkey_peft_f16.gguf",
-    "Q8_0": "leetmonkey_peft__q8_0.gguf",
     "Super Block Q6": "leetmonkey_peft_super_block_q6.gguf"
 }
 
-# Global dictionary to store loaded models
-loaded_models = {}
+def download_model(model_name):
+    logger.info(f"Downloading model: {model_name}")
+    model_path = hf_hub_download(
+        repo_id="sugiv/leetmonkey-peft-gguf",
+        filename=model_name,
+        cache_dir="./models",
+        force_download=True,
+        resume_download=True
+    )
+    logger.info(f"Model downloaded: {model_path}")
+    return model_path
+
+# Download the 8-bit model at startup
+q8_model_path = download_model(gguf_models["Q8_0"])
+
+# Load the 8-bit model
+llm = Llama(
+    model_path=q8_model_path,
+    n_ctx=2048,
+    n_threads=4,
+    n_gpu_layers=0,
+    verbose=False
+)
+
+logger.info("8-bit model loaded successfully")
 
-# Function to download and load the model
-def load_model(model_name):
-    model_path = hf_hub_download(repo_id="sugiv/leetmonkey-peft-gguf", filename=model_name)
-    return Llama(model_path=model_path, n_ctx=2048, n_threads=4, n_gpu_layers=0, verbose=False)
-
-# Function to preload all models
-def preload_models():
-    for name, file in gguf_models.items():
-        loaded_models[name] = load_model(file)
-    print("All models loaded successfully!")
-
-# Start preloading models in a separate thread
-threading.Thread(target=preload_models, daemon=True).start()
+# Load the dataset
+dataset = load_dataset("sugiv/leetmonkey_python_dataset")
+val_dataset = dataset["train"].train_test_split(test_size=0.1)["test"]
 
 # Generation parameters
 generation_kwargs = {
@@ -97,20 +114,16 @@ def extract_and_format_code(text):
     except:
         return formatted_code
 
-# Load the dataset
-dataset = load_dataset("sugiv/leetmonkey_python_dataset")
-val_dataset = dataset["train"].train_test_split(test_size=0.1)["test"]
-
 def update_problem():
     sample = random.choice(val_dataset)
     return sample['instruction']
 
-def update_solution(problem, model_name):
-    model = loaded_models.get(model_name)
-    if model is None:
-        return "Model is still loading. Please wait and try again."
-    generated_output = generate_solution(problem, model)
-    return extract_and_format_code(generated_output)
+def update_solution(problem):
+    logger.info("Generating solution")
+    generated_output = generate_solution(problem)
+    formatted_code = extract_and_format_code(generated_output)
+    logger.info("Solution generated successfully")
+    return formatted_code
 
 with gr.Blocks() as demo:
     gr.Markdown("# LeetCode Problem Solver")
@@ -121,11 +134,12 @@ with gr.Blocks() as demo:
             select_problem_btn = gr.Button("Select Random Problem")
         
         with gr.Column():
-            model_dropdown = gr.Dropdown(choices=list(gguf_models.keys()), label="Select GGUF Model", value="Exact Copy")
             solution_display = gr.Code(label="Generated Solution", language="python", lines=25)
             generate_btn = gr.Button("Generate Solution")
     
     select_problem_btn.click(update_problem, outputs=problem_display)
-    generate_btn.click(update_solution, inputs=[problem_display, model_dropdown], outputs=solution_display)
+    generate_btn.click(update_solution, inputs=[problem_display], outputs=solution_display)
 
-demo.launch(share=True)
+if __name__ == "__main__":
+    logger.info("Starting Gradio interface")
+    demo.launch(share=True)