Spaces:
Sleeping
Sleeping
File size: 13,238 Bytes
05a686e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 | """
Inference Script Example
===================================
MANDATORY
- Before submitting, ensure the following variables are defined in your environment configuration:
API_BASE_URL The API endpoint for the LLM.
MODEL_NAME The model identifier to use for inference.
HF_TOKEN Your Hugging Face / API key.
LOCAL_IMAGE_NAME The name of the local image to use for the environment if you are using from_docker_image()
method
- Defaults are set only for API_BASE_URL and MODEL_NAME
(and should reflect your active inference setup):
API_BASE_URL = os.getenv("API_BASE_URL", "<your-active-endpoint>")
MODEL_NAME = os.getenv("MODEL_NAME", "<your-active-model>")
- The inference script must be named `inference.py` and placed in the root directory of the project
- Participants must use OpenAI Client for all LLM calls using above variables
STDOUT FORMAT
- The script must emit exactly three line types to stdout, in this order:
[START] task=<task_name> env=<benchmark> model=<model_name>
[STEP] step=<n> action=<action_str> reward=<0.00> done=<true|false> error=<msg|null>
[END] success=<true|false> steps=<n> score=<score> rewards=<r1,r2,...,rn>
Rules:
- One [START] line at episode begin.
- One [STEP] line per step, immediately after env.step() returns.
- One [END] line after env.close(), always emitted (even on exception).
- reward and rewards are formatted to 2 decimal places.
- done and success are lowercase booleans: true or false.
- error is the raw last_action_error string, or null if none.
- All fields on a single line with no newlines within a line.
- Each tasks should return score in [0, 1]
Example:
[START] task=click-test env=miniwob model=Qwen3-VL-30B
[STEP] step=1 action=click('123') reward=0.00 done=false error=null
[STEP] step=2 action=fill('456','text') reward=0.00 done=false error=null
[STEP] step=3 action=click('789') reward=1.00 done=true error=null
[END] success=true steps=3 score=1.00 rewards=0.00,0.00,1.00
"""
import asyncio
import json
import os
import textwrap
from typing import Any, Callable, Dict, List, Optional
from openai import OpenAI
try:
from dotenv import load_dotenv
except ImportError:
load_dotenv = None
try:
from models import CoenvAction
from client import CoEnv
except ImportError:
from models import CoenvAction
from client import CoEnv
from server.graders.grader_pod_recovery import grade as grade_pod_recovery
from server.graders.grader_autoscaling import grade as grade_autoscaling
from server.graders.grader_incident import grade as grade_incident
if load_dotenv is not None:
load_dotenv()
LLM_BASE_URL = os.getenv("LLM_BASE_URL", "https://router.huggingface.co/v1")
ENV_URL = os.getenv("API_BASE_URL", "http://localhost:8000")
API_DELAY = float(os.getenv("API_DELAY", "0"))
MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen3-8B")
API_KEY = os.getenv("OPENROUTER_API_KEY") or os.getenv("HF_TOKEN")
BENCHMARKS = ["POD_RECOVERY", "AUTOSCALING", "INCIDENT"]
TASK_NAMES = ["pod_recovery", "autoscaling", "incident"]
TEMPERATURE = 0.7
MAX_TOKENS = 150
SUCCESS_SCORE_THRESHOLD = 0.1 # normalized score in [0, 1]
DEFAULT_MAX_STEPS = 15
SUCCESS_SCORE_THRESHOLD_BY_TASK: Dict[str, float] = {
"pod_recovery": 0.9,
"autoscaling": 0.9,
"incident": 0.8,
}
MAX_STALL_REPEATS = 4
REWARD_EPSILON = 1e-9
MAX_STEPS_BY_TASK = {
"pod_recovery": 15,
"autoscaling": 20,
"incident": 30,
}
GRADERS: Dict[str, Callable[[Dict[str, Any], int, int], float]] = {
"pod_recovery": grade_pod_recovery,
"autoscaling": grade_autoscaling,
"incident": grade_incident,
}
SYSTEM_PROMPT = textwrap.dedent(
"""
You are a Kubernetes incident-response agent.
Return ONLY valid JSON for one action with this schema:
{
"action_type": "scale|delete_pod|patch|rollout_restart|set_hpa|drain_node|describe|wait",
"deployment": "... optional ...",
"replicas": 1,
"pod_name": "...",
"resource_type": "deployment|pod|node|service|configmap|hpa",
"name": "...",
"patch": {},
"min_replicas": 1,
"max_replicas": 5,
"cpu_target_percent": 70,
"node_name": "..."
}
Do not include markdown, prose, or code fences.
"""
).strip()
def log_start(task: str, env: str, model: str) -> None:
print(f"[START] task={task} env={env} model={model}", flush=True)
def log_step(step: int, action: str, reward: float, done: bool, error: Optional[str]) -> None:
error_val = error if error else "null"
done_val = str(done).lower()
print(
f"[STEP] step={step} action={action} reward={reward:.2f} done={done_val} error={error_val}",
flush=True,
)
def log_end(success: bool, steps: int, score: float, rewards: List[float]) -> None:
rewards_str = ",".join(f"{r:.2f}" for r in rewards)
print(f"[END] success={str(success).lower()} steps={steps} score={score:.3f} rewards={rewards_str}", flush=True)
def _to_dict(obj: Any) -> Dict[str, Any]:
if hasattr(obj, "model_dump"):
return obj.model_dump()
if isinstance(obj, dict):
return obj
return vars(obj)
def _observation_summary(observation: Any) -> str:
obs = _to_dict(observation)
pods = obs.get("pods", [])
deployments = obs.get("deployments", [])
events = obs.get("events", [])
pod_status_counts: Dict[str, int] = {}
for pod in pods:
status = pod.get("status", "Unknown")
pod_status_counts[status] = pod_status_counts.get(status, 0) + 1
deployment_lines = []
for dep in deployments:
deployment_lines.append(
f"{dep.get('name')}: desired={dep.get('desired_replicas', 0)} available={dep.get('available_replicas', 0)}"
)
recent_events = [
f"{e.get('type', 'Normal')}/{e.get('reason', '')}: {e.get('message', '')}"
for e in events[-5:]
]
return textwrap.dedent(
f"""
Objective: {obs.get('objective', '')}
Step: {obs.get('step', 0)}
Pod status counts: {pod_status_counts}
Deployments:
{chr(10).join(deployment_lines) if deployment_lines else 'None'}
Recent events:
{chr(10).join(recent_events) if recent_events else 'None'}
"""
).strip()
def build_user_prompt(task_name: str, step: int, observation: Any, history: List[str]) -> str:
history_block = "\n".join(history[-4:]) if history else "None"
return textwrap.dedent(
f"""
Task: {task_name}
Step: {step}
Current cluster summary:
{_observation_summary(observation)}
Previous steps:
{history_block}
Return one valid next action as pure JSON.
"""
).strip()
def _safe_json_action(text: str) -> Optional[Dict[str, Any]]:
try:
return json.loads(text)
except json.JSONDecodeError:
start = text.find("{")
end = text.rfind("}")
if start != -1 and end != -1 and end > start:
try:
return json.loads(text[start : end + 1])
except json.JSONDecodeError:
return None
return None
def _heuristic_action(task_name: str, observation: Any) -> Dict[str, Any]:
obs = _to_dict(observation)
pods = obs.get("pods", [])
if task_name == "pod_recovery":
crashloop = [p for p in pods if p.get("deployment") == "frontend" and p.get("status") == "CrashLoopBackOff"]
if crashloop:
return {"action_type": "rollout_restart", "deployment": "frontend"}
return {"action_type": "describe", "resource_type": "deployment", "name": "frontend"}
if task_name == "autoscaling":
return {
"action_type": "set_hpa",
"deployment": "backend",
"min_replicas": 2,
"max_replicas": 6,
"cpu_target_percent": 70,
}
return {"action_type": "rollout_restart", "deployment": "auth-service"}
def _normalize_action(action: Dict[str, Any]) -> Dict[str, Any]:
action_type = action.get("action_type", "describe")
if isinstance(action_type, str):
action_type = {
"set_hpas": "set_hpa",
"hpa": "set_hpa",
"restart_rollout": "rollout_restart",
"noop": "wait",
"no_op": "wait",
"pause": "wait",
"sleep": "wait",
}.get(action_type.strip().lower(), action_type.strip().lower())
else:
action_type = "describe"
normalized: Dict[str, Any] = {"action_type": action_type}
allowed_fields = {
"deployment",
"replicas",
"pod_name",
"resource_type",
"name",
"patch",
"min_replicas",
"max_replicas",
"cpu_target_percent",
"node_name",
}
for field in allowed_fields:
if field in action and action[field] is not None:
normalized[field] = action[field]
defaults_by_type = {
"describe": {"resource_type": "deployment", "name": "frontend"},
"scale": {"deployment": "frontend", "replicas": 3},
"rollout_restart": {"deployment": "frontend"},
"delete_pod": {"pod_name": "frontend-unknown"},
"drain_node": {"node_name": "node-1"},
"patch": {"resource_type": "deployment", "name": "frontend", "patch": {}},
"set_hpa": {"deployment": "backend", "min_replicas": 2, "max_replicas": 6, "cpu_target_percent": 70},
"wait": {},
}
for k, v in defaults_by_type.get(action_type, {}).items():
normalized.setdefault(k, v)
return normalized
def get_model_action(client: OpenAI, task_name: str, step: int, observation: Any, history: List[str]) -> Dict[str, Any]:
user_prompt = build_user_prompt(task_name, step, observation, history)
try:
completion = client.chat.completions.create(
model=MODEL_NAME,
messages=[
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": user_prompt},
],
temperature=TEMPERATURE,
max_tokens=MAX_TOKENS,
stream=False,
)
text = (completion.choices[0].message.content or "").strip()
parsed = _safe_json_action(text)
if isinstance(parsed, dict):
return _normalize_action(parsed)
return _heuristic_action(task_name, observation)
except Exception as exc:
print(f"[DEBUG] Model request failed: {exc}", flush=True)
return _heuristic_action(task_name, observation)
async def main() -> None:
if not API_KEY:
raise RuntimeError("Missing HF_TOKEN/API_KEY for OpenAI client.")
for TASK_NAME, BENCHMARK in zip(TASK_NAMES, BENCHMARKS):
client = OpenAI(base_url=LLM_BASE_URL, api_key=API_KEY)
max_steps = MAX_STEPS_BY_TASK.get(TASK_NAME, DEFAULT_MAX_STEPS)
grader = GRADERS.get(TASK_NAME, grade_pod_recovery)
history: List[str] = []
rewards: List[float] = []
steps_taken = 0
score = 0.0
success = False
final_obs: Optional[Any] = None
episode_done = False
stalled = False
last_action_str: Optional[str] = None
consecutive_same_action = 0
last_reward: Optional[float] = None
log_start(task=TASK_NAME, env=BENCHMARK, model=MODEL_NAME)
try:
async with CoEnv.from_env("SandyTheAdventurer/coenv") as env:
result = await env.reset(task=TASK_NAME)
final_obs = result.observation
for step in range(1, max_steps + 1):
if API_DELAY > 0:
await asyncio.sleep(API_DELAY)
if result.done:
break
action_payload = get_model_action(client, TASK_NAME, step, final_obs, history)
action = CoenvAction(**action_payload)
result = await env.step(action)
obs = result.observation
final_obs = obs
reward = result.reward or 0.0
done = result.done
error = (obs.metadata or {}).get("error") if hasattr(obs, "metadata") else None
rewards.append(reward)
steps_taken = step
episode_done = bool(done)
action_str = json.dumps(action_payload, separators=(",", ":"))
log_step(step=step, action=action_str, reward=reward, done=done, error=error)
history.append(f"Step {step}: {action_str} -> reward {reward:+.2f}")
if done:
break
world_state = _to_dict(final_obs) if final_obs is not None else {}
score = grader(world_state, steps_taken, max_steps)
score = min(max(score, 0.0), 1.0)
success = (
episode_done
and not stalled
and steps_taken > 0
)
finally:
log_end(success=success, steps=steps_taken, score=score, rewards=rewards)
if __name__ == "__main__":
asyncio.run(main()) |