Spaces:
Sleeping
Sleeping
Commit ·
285e6b6
1
Parent(s): 63df97b
Use validator's API_KEY and API_BASE_URL from environment
Browse files- inference.py +29 -24
inference.py
CHANGED
|
@@ -18,10 +18,12 @@ except ImportError:
|
|
| 18 |
from environment.env import EmailTriageEnv
|
| 19 |
from environment.types import Action, EmailCategory, Team
|
| 20 |
|
| 21 |
-
# Environment variables
|
|
|
|
|
|
|
|
|
|
| 22 |
API_BASE_URL = os.getenv("API_BASE_URL", "https://api.openai.com/v1")
|
| 23 |
MODEL_NAME = os.getenv("MODEL_NAME", "gpt-4o-mini")
|
| 24 |
-
API_KEY = os.getenv("OPENAI_API_KEY")
|
| 25 |
|
| 26 |
# Configuration
|
| 27 |
MAX_STEPS = 50
|
|
@@ -200,9 +202,21 @@ def main() -> None:
|
|
| 200 |
tasks = ["spam_detection", "multi_class_routing", "context_aware_triage"]
|
| 201 |
all_scores = []
|
| 202 |
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
flush=True)
|
| 207 |
|
| 208 |
for task in tasks:
|
|
@@ -289,25 +303,16 @@ def main() -> None:
|
|
| 289 |
flush=True)
|
| 290 |
|
| 291 |
else:
|
| 292 |
-
# Normal mode: Use OpenAI API
|
| 293 |
-
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
|
| 298 |
-
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
|
| 302 |
-
for task in tasks:
|
| 303 |
-
try:
|
| 304 |
-
success, steps, score, rewards = run_task(client, task)
|
| 305 |
-
all_scores.append(score)
|
| 306 |
-
print(f"[TASK_SUMMARY] {task}: score={score:.3f} steps={steps}",
|
| 307 |
-
flush=True)
|
| 308 |
-
except Exception as e:
|
| 309 |
-
print(f"[TASK_ERROR] {task}: {e}", file=sys.stderr, flush=True)
|
| 310 |
-
all_scores.append(0.0)
|
| 311 |
|
| 312 |
# Final summary
|
| 313 |
avg_score = sum(all_scores) / len(all_scores) if all_scores else 0.0
|
|
|
|
| 18 |
from environment.env import EmailTriageEnv
|
| 19 |
from environment.types import Action, EmailCategory, Team
|
| 20 |
|
| 21 |
+
# Environment variables - check both formats
|
| 22 |
+
# Validator provides: API_KEY and API_BASE_URL
|
| 23 |
+
# Local usage: OPENAI_API_KEY
|
| 24 |
+
API_KEY = os.getenv("API_KEY") or os.getenv("OPENAI_API_KEY")
|
| 25 |
API_BASE_URL = os.getenv("API_BASE_URL", "https://api.openai.com/v1")
|
| 26 |
MODEL_NAME = os.getenv("MODEL_NAME", "gpt-4o-mini")
|
|
|
|
| 27 |
|
| 28 |
# Configuration
|
| 29 |
MAX_STEPS = 50
|
|
|
|
| 202 |
tasks = ["spam_detection", "multi_class_routing", "context_aware_triage"]
|
| 203 |
all_scores = []
|
| 204 |
|
| 205 |
+
# Try to initialize OpenAI client if API key is available
|
| 206 |
+
client = None
|
| 207 |
+
if API_KEY and OPENAI_AVAILABLE:
|
| 208 |
+
try:
|
| 209 |
+
# Initialize with validator's provided API_BASE_URL and API_KEY
|
| 210 |
+
client = OpenAI(api_key=API_KEY, base_url=API_BASE_URL)
|
| 211 |
+
print(f"[INFO] Using API endpoint: {API_BASE_URL}", flush=True)
|
| 212 |
+
except Exception as e:
|
| 213 |
+
print(f"[WARNING] Failed to initialize OpenAI client: {e}",
|
| 214 |
+
file=sys.stderr, flush=True)
|
| 215 |
+
client = None
|
| 216 |
+
|
| 217 |
+
if client is None:
|
| 218 |
+
# Demo/Validation mode: No API key or OpenAI not available
|
| 219 |
+
print("[WARNING] No API credentials available. Running in validation mode.",
|
| 220 |
flush=True)
|
| 221 |
|
| 222 |
for task in tasks:
|
|
|
|
| 303 |
flush=True)
|
| 304 |
|
| 305 |
else:
|
| 306 |
+
# Normal mode: Use OpenAI API (through validator's proxy if available)
|
| 307 |
+
for task in tasks:
|
| 308 |
+
try:
|
| 309 |
+
success, steps, score, rewards = run_task(client, task)
|
| 310 |
+
all_scores.append(score)
|
| 311 |
+
print(f"[TASK_SUMMARY] {task}: score={score:.3f} steps={steps}",
|
| 312 |
+
flush=True)
|
| 313 |
+
except Exception as e:
|
| 314 |
+
print(f"[TASK_ERROR] {task}: {e}", file=sys.stderr, flush=True)
|
| 315 |
+
all_scores.append(0.0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 316 |
|
| 317 |
# Final summary
|
| 318 |
avg_score = sum(all_scores) / len(all_scores) if all_scores else 0.0
|