Update backend.py

backend.py

@@ -1,4 +1,4 @@
-# backend.py — FINAL HARDENED VERSION v1.1
+# backend.py — FINAL HARDENED VERSION v1.2
 import sqlite3
 import os
 import json
@@ -74,7 +74,6 @@ def load_model(model_name):
     if model_name in _MODEL_CACHE: return _MODEL_CACHE[model_name]
     print(f"Loading model: {model_name}...")
     tokenizer = AutoTokenizer.from_pretrained(model_name)
-    # --- THIS IS THE FIX FOR THE WARNING ---
     model = AutoModelForCausalLM.from_pretrained(model_name, dtype="auto", device_map="auto", trust_remote_code=True, attn_implementation="eager")
     _MODEL_CACHE[model_name] = (tokenizer, model)
     print(f"Model {model_name} loaded and cached.")
@@ -82,8 +81,8 @@ def load_model(model_name):
 
 # ------------------------------ AGENT PROMPTS (SIMPLIFIED & ROBUST) ------------------------------
 ROLE_PROMPTS = {
-    "planner": """You are an expert software planner. Break down the user's request into a detailed plan.
-    Output ONLY a single JSON object with the keys: "purpose", "features", "dependencies", and "files".
+    "planner": """You are an expert file planner. Based on the user's request, determine the necessary file structure.
+    Output ONLY a single JSON object with a single key: "files".
     The "files" key MUST be an array of strings representing complete file paths (e.g., ["src/main.py", "tests/test_main.py", "requirements.txt"]).""",
     "architect": """You are a software architect. Create initial placeholder content for a list of files.
     Output ONLY a single JSON object where keys are file paths and values are the initial content (e.g., a comment like '# Main application logic here').""",
@@ -135,13 +134,15 @@ def generate_with_model(role: str, prompt: str) -> str:
         messages = [{"role": "system", "content": ROLE_PROMPTS[role]}, {"role": "user", "content": prompt}]
         input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
         inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
-        outputs = model.generate(**inputs, max_new_tokens=2048, pad_token_id=tokenizer.eos_token_id)
+        
+        outputs = model.generate(**inputs, max_new_tokens=2048, pad_token_id=tokenizer.eos_token_id, use_cache=False)
+        
         return tokenizer.decode(outputs[0][len(inputs.input_ids[0]):], skip_special_tokens=True).strip()
     except Exception as e:
         print(f"Error during model generation for role {role}: {e}")
         return f'{{"error": "Failed to generate response: {str(e)}"}}'
 
-# ------------------------------ THE AGENT CHAIN EXECUTOR (REWRITTEN FOR RELIABILITY) ------------------------------
+# ------------------------------ THE AGENT CHAIN EXECUTOR ------------------------------
 def run_agent_chain(project_id, user_id, initial_prompt):
     project_dir = get_project_dir(user_id, project_id)
     log_entries = []
@@ -156,16 +157,12 @@ def run_agent_chain(project_id, user_id, initial_prompt):
         log_step("SYSTEM", "Initializing project...")
 
         # 1. PLANNER
-        log_step("PLANNER", "Analyzing user request and creating a project plan...")
+        log_step("PLANNER", "Determining file structure from user request...")
         plan_response = generate_with_model("planner", initial_prompt)
         plan_data = _extract_json(plan_response)
         if not plan_data or "files" not in plan_data: raise ValueError("Planner failed to create a valid JSON plan with a 'files' key.")
         
-        if plan_data["files"] and isinstance(plan_data["files"][0], dict):
-            log_step("SYSTEM", "Correcting malformed file list from Planner...")
-            plan_data["files"] = [item["file"] for item in plan_data["files"] if "file" in item]
-
-        log_step("PLANNER", "Plan created.", json.dumps(plan_data, indent=2))
+        log_step("PLANNER", "File structure plan created.", json.dumps(plan_data, indent=2))
         
         # 2. ARCHITECT
         log_step("ARCHITECT", "Creating initial file skeletons...")
@@ -180,7 +177,7 @@ def run_agent_chain(project_id, user_id, initial_prompt):
         source_files = [f for f in plan_data['files'] if f.startswith('src/') and f.endswith('.py')]
         for file_path in source_files:
             log_step("CODER", f"Writing complete code for `{file_path}`...")
-            coder_prompt = f"Project purpose: {plan_data['purpose']}. Write the full Python code for the file: `{file_path}`."
+            coder_prompt = f"Based on the user's overall request: '{initial_prompt}'.\n\nWrite the full Python code for the file: `{file_path}`."
             code = generate_with_model("coder", coder_prompt)
             create_file(project_dir, file_path, code)
             log_step("CODER", f"Finished writing `{file_path}`.", code)