Upload 4 files

app.py

@@ -8,13 +8,13 @@ def process_error(error_message: str) -> str:
     """
     Main function to process the error message and return JSON response.
     """
-    # Build full prompt
-    full_prompt = SYSTEM_PROMPT + error_message.strip()
+    # Build full prompt with proper ending
+    full_prompt = SYSTEM_PROMPT + error_message.strip() + "\n<|end|>\n<|assistant|>"
     
     # Get raw model output
     raw_output = generate_response(full_prompt)
     
-    # Extrahiere JSON aus der Antwort (falls Modell Prompt wiederholt)
+    # Extrahiere JSON aus der Antwort
     json_match = re.search(r"\{.*\}", raw_output, re.DOTALL)
     if json_match:
         raw_output = json_match.group(0)

model.py

@@ -2,15 +2,12 @@ import os
 from dotenv import load_dotenv
 from transformers import pipeline
 import torch
+from model_config import MODEL_NAME
 
 # Load environment variables from .env file
 load_dotenv()
 
-# Initialize the model pipeline (adjust model name as needed)
-# Using a small, CPU-friendly model like StarCoder-1B
-MODEL_NAME = "bigcode/starcoderbase-1b"
-
-# Get HuggingFace token from environment variable
+# Get HuggingFace token from environment variable (for gated models only)
 HF_TOKEN = os.getenv("HF_TOKEN")
 
 # Initialize text generation pipeline
@@ -20,7 +17,7 @@ generator = pipeline(
     model=MODEL_NAME,
     torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
     device_map="auto",  # Automatically choose device (CPU if no GPU)
-    token=HF_TOKEN      # Pass token for gated models
+    token=HF_TOKEN if "bigcode" in MODEL_NAME else None  # Only for gated repos
 )
 
 def generate_response(prompt: str) -> str:

model_config.py

@@ -0,0 +1,3 @@
+# Modell-Konfiguration für CPU-gestützte Inferenz
+
+MODEL_NAME = "microsoft/Phi-3-mini-4k-instruct"  # Öffentliches, schnelles Modell

prompt.py

@@ -1,15 +1,17 @@
 # System prompt to instruct the model to always respond in structured JSON format
 
 SYSTEM_PROMPT = """
+<|system|>
 You are a helpful assistant for developers debugging code.
 The user will provide a stack trace or error message.
 Your task is to provide a JSON object with exactly two keys:
 - "explanation": A concise, understandable explanation of the error (max 2 sentences)
 - "fix_suggestion": A clear, actionable fix in code or concept (max 3 sentences)
 
-Respond ONLY with valid JSON. No other text. No markdown. No explanations.
+Respond ONLY with valid JSON. No other text. No markdown.
 Example:
 {"explanation": "Variable is undefined.", "fix_suggestion": "Check if defined."}
+<|end|>
 
-Input:
+<|user|>
 """