Update coordinator/task_parser.py

coordinator/task_parser.py

@@ -2,34 +2,30 @@
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 
-# Use a larger model for better reasoning
-MODEL_NAME = "facebook/opt-1.3b"
+# Use a larger OPT model if possible for better reasoning
+MODEL_NAME = "facebook/opt-1.3b"  # upgrade from 125m
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
 
 def parse_brief_with_reasoning(brief: str):
     """
-    Use LLM to extract tasks and reasoning
+    Parses a project brief into distinct tasks with LLM reasoning.
     Returns a dict:
         {
-            "reasoning": <llm reasoning>,
+            "reasoning": <str>,
             "tasks": [list of tasks]
         }
     """
     prompt = f"""
-You are an expert project manager. Given the project brief, list the distinct technical tasks needed to implement it.
-Explain your reasoning step by step.
-Output format:
-Reasoning:
-- Explain step 1
-- Explain step 2
-Tasks:
-1. First task
-2. Second task
+You are an expert project manager. 
+Given the project brief, break it into **distinct technical tasks**, each on a new line, numbered.
+Explain your reasoning step by step before listing the tasks.
 Do not repeat the project brief.
 
 Project Brief:
 {brief}
+
+Reasoning and Tasks:
 """
     inputs = tokenizer(prompt, return_tensors="pt")
     outputs = model.generate(**inputs, max_new_tokens=300)
@@ -44,7 +40,7 @@ Project Brief:
     # Deduplicate tasks
     tasks = []
     for line in tasks_text.split("\n"):
-        task = line.strip("-123. ").strip()
+        task = line.strip("- ").strip()
         if task and task not in tasks:
             tasks.append(task)