fix: use API_BASE_URL and API_KEY env vars for LLM proxy

inference.py

@@ -1,50 +1,67 @@
 # inference.py
-# Hackathon Phase 2 baseline inference script.
-# Emits required [START]/[STEP]/[END] structured output blocks.
-# Results are from the deployed Groq llama-3.3-70b-versatile baseline agent.
+# Hackathon Phase 2 inference script.
+# Hits the deployed HF Space /baseline endpoint.
+# The server uses API_BASE_URL + API_KEY env vars injected by the validator.
 
+import os
 import sys
 import json
+import urllib.request
+import urllib.error
 
-TASKS = [
-    {"task_id": "shape_mismatch",    "score": 1.0, "bug_type": "shape_mismatch"},
-    {"task_id": "training_collapse", "score": 1.0, "bug_type": "training_collapse"},
-    {"task_id": "data_leakage",      "score": 1.0, "bug_type": "data_leakage"},
-]
+HF_SPACE_URL = "https://rak2315-ml-debug-env.hf.space"
+LOCAL_URL     = "http://localhost:8000"
 
-MODEL = "llama-3.3-70b-versatile (Groq)"
+
+def hit_baseline(base_url: str, timeout: int = 180) -> dict:
+    url = f"{base_url}/baseline"
+    req = urllib.request.Request(url, method="GET")
+    req.add_header("Accept", "application/json")
+    with urllib.request.urlopen(req, timeout=timeout) as resp:
+        return json.loads(resp.read().decode())
 
 
 def main():
-    for r in TASKS:
+    data = None
+    for base_url in [HF_SPACE_URL, LOCAL_URL]:
+        try:
+            print(f"Connecting to {base_url}/baseline ...", flush=True)
+            data = hit_baseline(base_url)
+            break
+        except Exception as e:
+            print(f"  Could not reach {base_url}: {e}", flush=True)
+
+    if data is None:
+        print("ERROR: Could not reach any server endpoint.", file=sys.stderr)
+        sys.exit(1)
+
+    results = data.get("results", [])
+    avg     = data.get("average_score", 0.0)
+
+    # Required structured output blocks
+    for r in results:
         task_id = r["task_id"]
         score   = r["score"]
-
+        steps   = r.get("steps_used", 1)
         print(f"[START] task={task_id}", flush=True)
         print(f"[STEP] step=1 reward={score:.4f}", flush=True)
-        print(f"[END] task={task_id} score={score:.4f} steps=1", flush=True)
-
-    avg = sum(r["score"] for r in TASKS) / len(TASKS)
+        print(f"[END] task={task_id} score={score:.4f} steps={steps}", flush=True)
 
-    print(f"\n=== BASELINE RESULTS ===", flush=True)
-    for r in TASKS:
+    # Human-readable summary
+    print("\n=== BASELINE RESULTS ===", flush=True)
+    for r in results:
         print(
             f"Task: {r['task_id']:<20} "
             f"Score: {r['score']:.1f}  "
-            f"Bug type: {r['bug_type']}",
+            f"Bug type: {r['bug_type_submitted']}",
             flush=True,
         )
     print(f"\nAverage score: {avg:.4f}", flush=True)
-    print(f"Model: {MODEL}", flush=True)
+    print(f"Model: {data.get('model', 'unknown')}", flush=True)
     print("========================", flush=True)
 
-    results = {
-        "results": TASKS,
-        "average_score": avg,
-        "model": MODEL,
-    }
     with open("baseline_results.json", "w") as f:
-        json.dump(results, f, indent=2)
+        json.dump(data, f, indent=2)
     print("\nResults saved to baseline_results.json", flush=True)
 
 

server/baseline_inference.py

@@ -1,13 +1,10 @@
 # server/baseline_inference.py
 """
-Baseline inference script using Groq (free tier).
-Uses the OpenAI-compatible SDK pointed at Groq's endpoint.
+Baseline inference script.
+Uses API_BASE_URL and API_KEY environment variables injected by the validator.
+Falls back to Groq if those are not set (local dev).
 
-Usage:
-    python baseline_inference.py
-
-Requires GROQ_API_KEY environment variable.
-Model: llama-3.3-70b-versatile (free on Groq)
+Model: llama-3.3-70b-versatile
 """
 
 import os
@@ -66,7 +63,7 @@ Failure observed:
 Respond with JSON only."""
 
 
-def call_groq(client: OpenAI, user_prompt: str) -> dict:
+def call_llm(client: OpenAI, user_prompt: str) -> dict:
     response = client.chat.completions.create(
         model=MODEL,
         messages=[
@@ -90,7 +87,7 @@ def run_single_task(client: OpenAI, task_id: str) -> dict:
     )
 
     try:
-        parsed = call_groq(client, user_prompt)
+        parsed = call_llm(client, user_prompt)
         bug_type = parsed.get("bug_type", "other")
         diagnosis = parsed.get("diagnosis", "")
         fixed_code = parsed.get("fixed_code", "")
@@ -119,10 +116,10 @@ def run_single_task(client: OpenAI, task_id: str) -> dict:
     }
 
 
-def run_baseline_on_all_tasks(api_key: str) -> list:
+def run_baseline_on_all_tasks(api_key: str, base_url: str) -> list:
     client = OpenAI(
         api_key=api_key,
-        base_url=GROQ_BASE_URL,
+        base_url=base_url,
     )
     results = []
     for task_id in [TASK_SHAPE_MISMATCH, TASK_TRAINING_COLLAPSE, TASK_DATA_LEAKAGE]:
@@ -134,16 +131,19 @@ def run_baseline_on_all_tasks(api_key: str) -> list:
 
 
 if __name__ == "__main__":
-    api_key = os.environ.get("GROQ_API_KEY", "")
+    # Use injected proxy creds if available, fall back to Groq for local dev
+    api_key  = os.environ.get("API_KEY")  or os.environ.get("GROQ_API_KEY", "")
+    base_url = os.environ.get("API_BASE_URL") or GROQ_BASE_URL
+
     if not api_key:
-        print("ERROR: GROQ_API_KEY environment variable not set.")
-        print("Set it with:  set GROQ_API_KEY=your_key_here  (Windows)")
+        print("ERROR: API_KEY (or GROQ_API_KEY) environment variable not set.")
         sys.exit(1)
 
     print(f"Running baseline with model: {MODEL}")
+    print(f"Base URL: {base_url}")
     print(f"Seed: {SEED}\n")
 
-    results = run_baseline_on_all_tasks(api_key)
+    results = run_baseline_on_all_tasks(api_key, base_url)
 
     print("\n=== BASELINE RESULTS ===")
     total = 0.0