Spaces:
Sleeping
Sleeping
Commit ·
804fa70
1
Parent(s): 8828035
fix: refactor baseline evaluation llm driver to enforce openenv strict pre-submission checklist constraints including dynamic api variables and standard stdout lifecycle tags
Browse files- baseline/run_baseline.py +14 -7
baseline/run_baseline.py
CHANGED
|
@@ -3,16 +3,21 @@ import sys
|
|
| 3 |
import time
|
| 4 |
import json
|
| 5 |
import httpx
|
| 6 |
-
from
|
| 7 |
|
| 8 |
def main():
|
| 9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
if not api_key:
|
| 11 |
-
print("ERROR:
|
| 12 |
sys.exit(1)
|
| 13 |
|
|
|
|
| 14 |
base_url = os.environ.get("DATAENGENV_URL", "http://localhost:8000")
|
| 15 |
-
client = Groq(api_key=api_key)
|
| 16 |
|
| 17 |
system_prompt = """You are a Data Engineer debugging a broken ML pipeline.
|
| 18 |
You have a MAXIMUM of 15 steps. Use them wisely.
|
|
@@ -75,8 +80,9 @@ Read them carefully before deciding your action.
|
|
| 75 |
"role": "user",
|
| 76 |
"content": f"Task started. Script:\n{obs.get('script_content', '')}\n\nError: {obs.get('last_run_error', 'none')}"
|
| 77 |
})
|
| 78 |
-
|
| 79 |
for step in range(15):
|
|
|
|
| 80 |
if done:
|
| 81 |
break
|
| 82 |
|
|
@@ -131,7 +137,7 @@ Read them carefully before deciding your action.
|
|
| 131 |
try:
|
| 132 |
trimmed = [messages[0]] + messages[-6:]
|
| 133 |
resp = client.chat.completions.create(
|
| 134 |
-
model=
|
| 135 |
messages=trimmed,
|
| 136 |
response_format={"type": "json_object"},
|
| 137 |
temperature=0.0
|
|
@@ -139,7 +145,7 @@ Read them carefully before deciding your action.
|
|
| 139 |
action_str = resp.choices[0].message.content
|
| 140 |
break
|
| 141 |
except Exception as e:
|
| 142 |
-
print(f"
|
| 143 |
if attempt == 1:
|
| 144 |
print("Skipping step due to repeated API failures.")
|
| 145 |
|
|
@@ -186,6 +192,7 @@ Read them carefully before deciding your action.
|
|
| 186 |
f"Steps remaining: {15 - step - 1}"
|
| 187 |
})
|
| 188 |
|
|
|
|
| 189 |
try:
|
| 190 |
g_res = http.post(f"{base_url}/grader", json={"task_id": task_id})
|
| 191 |
g_res.raise_for_status()
|
|
|
|
| 3 |
import time
|
| 4 |
import json
|
| 5 |
import httpx
|
| 6 |
+
from openai import OpenAI
|
| 7 |
|
| 8 |
def main():
|
| 9 |
+
API_BASE_URL = os.getenv("API_BASE_URL", "https://api.groq.com/openai/v1")
|
| 10 |
+
MODEL_NAME = os.getenv("MODEL_NAME", "llama-3.1-8b-instant")
|
| 11 |
+
HF_TOKEN = os.getenv("HF_TOKEN")
|
| 12 |
+
LOCAL_IMAGE_NAME = os.getenv("LOCAL_IMAGE_NAME")
|
| 13 |
+
|
| 14 |
+
api_key = HF_TOKEN if HF_TOKEN else os.environ.get("GROQ_API_KEY")
|
| 15 |
if not api_key:
|
| 16 |
+
print("ERROR: API key not found. Set HF_TOKEN or GROQ_API_KEY")
|
| 17 |
sys.exit(1)
|
| 18 |
|
| 19 |
+
client = OpenAI(base_url=API_BASE_URL, api_key=api_key)
|
| 20 |
base_url = os.environ.get("DATAENGENV_URL", "http://localhost:8000")
|
|
|
|
| 21 |
|
| 22 |
system_prompt = """You are a Data Engineer debugging a broken ML pipeline.
|
| 23 |
You have a MAXIMUM of 15 steps. Use them wisely.
|
|
|
|
| 80 |
"role": "user",
|
| 81 |
"content": f"Task started. Script:\n{obs.get('script_content', '')}\n\nError: {obs.get('last_run_error', 'none')}"
|
| 82 |
})
|
| 83 |
+
print("START")
|
| 84 |
for step in range(15):
|
| 85 |
+
print("STEP")
|
| 86 |
if done:
|
| 87 |
break
|
| 88 |
|
|
|
|
| 137 |
try:
|
| 138 |
trimmed = [messages[0]] + messages[-6:]
|
| 139 |
resp = client.chat.completions.create(
|
| 140 |
+
model=MODEL_NAME,
|
| 141 |
messages=trimmed,
|
| 142 |
response_format={"type": "json_object"},
|
| 143 |
temperature=0.0
|
|
|
|
| 145 |
action_str = resp.choices[0].message.content
|
| 146 |
break
|
| 147 |
except Exception as e:
|
| 148 |
+
print(f"API failed (attempt {attempt+1}): {e}")
|
| 149 |
if attempt == 1:
|
| 150 |
print("Skipping step due to repeated API failures.")
|
| 151 |
|
|
|
|
| 192 |
f"Steps remaining: {15 - step - 1}"
|
| 193 |
})
|
| 194 |
|
| 195 |
+
print("END")
|
| 196 |
try:
|
| 197 |
g_res = http.post(f"{base_url}/grader", json={"task_id": task_id})
|
| 198 |
g_res.raise_for_status()
|