final fix: HF server

inference.py

@@ -1,22 +1,20 @@
 """
 inference.py — AI Code Review OpenEnv Environment
 Entry point for the hackathon validator.
-
-Log format (strict):
-  [START] task=... env=... model=...
-  [STEP] step=N action=... reward=0.00 done=true|false error=null|msg
-  [END] success=true|false steps=N rewards=r1,r2,...
 """
 
 import os
 import sys
+import time
+import threading
+from http.server import HTTPServer, BaseHTTPRequestHandler
 
 from openai import OpenAI
 from env.environment import CodeReviewEnvironment
 from env.models import Action
 
 # ---------------------------------------------------------------------------
-# Environment variables (mandatory / with defaults)
+# Environment variables
 # ---------------------------------------------------------------------------
 API_BASE_URL = os.getenv("API_BASE_URL", "https://api.openai.com/v1")
 MODEL_NAME   = os.getenv("MODEL_NAME",   "gpt-4.1-mini")
@@ -25,6 +23,20 @@ HF_TOKEN     = os.getenv("HF_TOKEN")
 ENV_NAME  = "ai-code-review-env"
 TASK_NAME = "code-review"
 
+# ---------------------------------------------------------------------------
+# Simple HTTP server (REQUIRED for Hugging Face Spaces)
+# ---------------------------------------------------------------------------
+
+class SimpleHandler(BaseHTTPRequestHandler):
+    def do_GET(self):
+        self.send_response(200)
+        self.end_headers()
+        self.wfile.write(b"OK")
+
+def start_server():
+    server = HTTPServer(("0.0.0.0", 7860), SimpleHandler)
+    server.serve_forever()
+
 # ---------------------------------------------------------------------------
 # Prompt builders
 # ---------------------------------------------------------------------------
@@ -35,7 +47,6 @@ SYSTEM_PROMPT = """You are an expert code reviewer. You will be shown buggy code
 
 Be concise and precise. When fixing, output the complete corrected function wrapped in ```javascript ... ```."""
 
-
 def build_identify_prompt(obs) -> str:
     return (
         f"Task: {obs.task}\n\n"
@@ -43,7 +54,6 @@ def build_identify_prompt(obs) -> str:
         f"IDENTIFY the bug: explain what is wrong and why."
     )
 
-
 def build_fix_prompt(obs, identify_response: str) -> str:
     return (
         f"Task: {obs.task}\n\n"
@@ -52,30 +62,27 @@ def build_fix_prompt(obs, identify_response: str) -> str:
         f"FIX the code: return ONLY the corrected code in a ```{obs.language} ... ``` block."
     )
 
-
 # ---------------------------------------------------------------------------
 # LLM call
 # ---------------------------------------------------------------------------
 
 def call_llm(client: OpenAI, messages: list) -> str:
-    """Call the LLM and return the text response. Raises on failure."""
     response = client.chat.completions.create(
         model=MODEL_NAME,
         messages=messages,
         max_tokens=512,
-        temperature=0.0,   # deterministic
+        temperature=0.0,
     )
-    return response.choices[0].message.content.strip()
-
+    content = response.choices[0].message.content or ""
+    return content.strip()
 
 # ---------------------------------------------------------------------------
-# Log helpers — strict format, no deviation
+# Logging helpers
 # ---------------------------------------------------------------------------
 
 def log_start():
     print(f"[START] task={TASK_NAME} env={ENV_NAME} model={MODEL_NAME}", flush=True)
 
-
 def log_step(step: int, action: str, reward: float, done: bool, error=None):
     error_str = "null" if error is None else str(error).replace("\n", " ")
     done_str  = "true" if done else "false"
@@ -85,19 +92,16 @@ def log_step(step: int, action: str, reward: float, done: bool, error=None):
         flush=True,
     )
 
-
 def log_end(success: bool, steps: int, rewards: list):
     success_str  = "true" if success else "false"
     rewards_str  = ",".join(f"{r:.2f}" for r in rewards)
     print(f"[END] success={success_str} steps={steps} rewards={rewards_str}", flush=True)
 
-
 # ---------------------------------------------------------------------------
-# Main episode loop
+# Main logic
 # ---------------------------------------------------------------------------
 
 def run():
-    # Validate mandatory env var
     if not HF_TOKEN:
         log_start()
         log_end(success=False, steps=0, rewards=[])
@@ -109,63 +113,58 @@ def run():
 
     log_start()
 
-    step_num      = 0
-    rewards       = []
-    success       = False
-    conversation  = []   # rolling message history for the LLM
+    step_num = 0
+    rewards  = []
+    success  = False
 
     try:
         obs = env.reset()
 
         while True:
-            task = env.state
-
-            # ── IDENTIFY step ────────────────────────────────────────────
+            # IDENTIFY
             step_num += 1
             error = None
-            identify_text = ""
 
             try:
-                conversation = [
+                messages = [
                     {"role": "system", "content": SYSTEM_PROMPT},
-                    {"role": "user",   "content": build_identify_prompt(obs)},
+                    {"role": "user", "content": build_identify_prompt(obs)},
                 ]
-                identify_text = call_llm(client, conversation)
+                identify_text = call_llm(client, messages)
             except Exception as e:
+                identify_text = ""
                 error = f"LLM error: {e}"
 
-            identify_action = f"identify:task{obs.task_id}"
             result = env.step(Action(action_type="identify", content=identify_text))
-
             reward = result["reward"]
             done   = result["done"]
+
             rewards.append(reward)
-            log_step(step_num, identify_action, reward, done, error)
+            log_step(step_num, f"identify:task{obs.task_id}", reward, done, error)
 
             if done:
                 break
 
             obs = result["observation"]
 
-            # ── FIX step ─────────────────────────────────────────────────
+            # FIX
             step_num += 1
-            error    = None
-            fix_text = ""
+            error = None
 
             try:
-                conversation.append({"role": "assistant", "content": identify_text})
-                conversation.append({"role": "user",      "content": build_fix_prompt(obs, identify_text)})
-                fix_text = call_llm(client, conversation)
+                messages.append({"role": "assistant", "content": identify_text})
+                messages.append({"role": "user", "content": build_fix_prompt(obs, identify_text)})
+                fix_text = call_llm(client, messages)
             except Exception as e:
+                fix_text = ""
                 error = f"LLM error: {e}"
 
-            fix_action = f"fix:task{obs.task_id}"
             result = env.step(Action(action_type="fix", content=fix_text))
-
             reward = result["reward"]
             done   = result["done"]
+
             rewards.append(reward)
-            log_step(step_num, fix_action, reward, done, error)
+            log_step(step_num, f"fix:task{obs.task_id}", reward, done, error)
 
             if done:
                 break
@@ -175,26 +174,25 @@ def run():
         success = True
 
     except Exception as e:
-        # Catch-all — always print END
-        log_step(step_num + 1, "error", 0.00, True, error=str(e))
         log_end(success=False, steps=step_num, rewards=rewards)
-        sys.exit(1)
+        print(f"[FATAL] {e}", flush=True)
+        return
 
     log_end(success=success, steps=step_num, rewards=rewards)
 
-
 # ---------------------------------------------------------------------------
 # Entry point
 # ---------------------------------------------------------------------------
 
-import time
-
 if __name__ == "__main__":
+    # Start HTTP server (required for HF Spaces)
+    threading.Thread(target=start_server, daemon=True).start()
+
     try:
         run()
     except Exception as e:
         print(f"[FATAL] {e}", flush=True)
 
-    # Keep container alive for HF Spaces
+    # Keep container alive
     while True:
         time.sleep(60)