File size: 10,399 Bytes
cc6e5ac | 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 | """
Inference Script — TICE (Tumor Immune Control Environment)
=========================================================
MANDATORY ENVIRONMENT VARIABLES:
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 Docker image name for the environment.
STDOUT FORMAT (strictly followed):
[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>
This script runs 3 tasks (easy, medium, hard). Each task is a single multi-step episode:
reset() → repeatedly: LLM picks (B-cell action, T-cell action) → step() → log → done
Final score per task = average reward across steps in that episode.
Overall score = average across all 3 tasks.
"""
from __future__ import annotations
import asyncio
import json
import os
import textwrap
from pathlib import Path
from typing import Any, List, Optional, Tuple
from dotenv import load_dotenv
from openai import OpenAI
try:
from tice import TICEAction, TICEEnv
from tice.models import B_CELL_ACTIONS, T_CELL_ACTIONS
except (ImportError, ModuleNotFoundError):
from client import TICEEnv
from models import B_CELL_ACTIONS, TICEAction, T_CELL_ACTIONS
# Load .env before reading env vars
load_dotenv(Path(__file__).resolve().parent / ".env")
# --- Config (match judging expectations) ---
IMAGE_NAME = os.getenv("LOCAL_IMAGE_NAME") or os.getenv("IMAGE_NAME")
API_KEY = os.getenv("HF_TOKEN") or os.getenv("API_KEY") or os.getenv("OPENAI_API_KEY")
API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")
MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")
BENCHMARK = "tice"
TEMPERATURE = float(os.getenv("TICE_LLM_TEMPERATURE", "0.2"))
MAX_TOKENS = int(os.getenv("TICE_LLM_MAX_TOKENS", "500"))
SUCCESS_SCORE_THRESHOLD = float(os.getenv("TICE_SUCCESS_SCORE_THRESHOLD", "0.0"))
TASKS: List[Tuple[str, str, str]] = [
("easy", "immune_cold", "easy"),
("medium", "immune_hot", "medium"),
("hard", "high_mutation", "hard"),
]
SYSTEM_PROMPT = textwrap.dedent(
f"""
You control a tumor immune therapy simulator. On each turn you must choose exactly one
B-cell action and one T-cell action.
Valid B-cell actions:
{", ".join(B_CELL_ACTIONS)}
Valid T-cell actions:
{", ".join(T_CELL_ACTIONS)}
Objective:
- Reduce and eradicate the tumor before timeout.
- Preserve energy and avoid excessive B-cell and T-cell fatigue.
- B cells improve detection. T cells do the damage.
- In early phase, overcommitting T cells before reliable detection is usually wasteful.
- If T-cell fatigue is high, recovery may be better than aggression.
You must reply with a valid JSON object and nothing else:
{{
"b_cell_action": "<one valid B-cell action>",
"t_cell_action": "<one valid T-cell action>",
"reasoning": "<brief reasoning>"
}}
Rules:
- Use only the valid action strings.
- Base decisions only on the provided observation.
- Keep reasoning short.
"""
).strip()
# ---------------------------------------------------------------------------
# Logging helpers — exact format required by hackathon judges
# ---------------------------------------------------------------------------
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()
action_clean = action.replace("\n", " ").replace("\r", "")[:120]
print(
f"[STEP] step={step} action={action_clean} reward={reward:.2f} "
f"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} "
f"score={score:.3f} rewards={rewards_str}",
flush=True,
)
def heuristic_action(observation) -> tuple[str, str]:
if observation.episode_phase == "early":
return "INCREASE_HIGH", "REST"
if observation.detection_signal < 0.4:
return "INCREASE_LOW", "ATTACK_LOW"
if observation.t_cell_fatigue > 0.6:
return "MAINTAIN", "REST"
if observation.tumor_trend == "increasing":
return "MAINTAIN", "ATTACK_MEDIUM"
return "MAINTAIN", "ATTACK_LOW"
def build_user_prompt(observation) -> str:
return textwrap.dedent(
f"""
Current episode context:
- archetype: {observation.archetype}
- difficulty: {observation.difficulty}
- timestep: {observation.timestep}
- episode_phase: {observation.episode_phase}
- tumor_trend: {observation.tumor_trend}
- detection_signal: {observation.detection_signal}
- t_cell_effectiveness: {observation.t_cell_effectiveness}
- resource_level: {observation.resource_level}
- b_cell_fatigue: {observation.b_cell_fatigue}
- t_cell_fatigue: {observation.t_cell_fatigue}
- recent_outcome: {observation.recent_outcome}
- feedback: {observation.feedback}
Choose the next B-cell and T-cell actions.
Respond with JSON only.
"""
).strip()
def sanitize_json_response(raw_response: str) -> str:
cleaned = raw_response.strip()
if cleaned.startswith("```"):
lines = cleaned.splitlines()
cleaned = "\n".join(
line for line in lines if not line.strip().startswith("```")
).strip()
return cleaned
def coerce_action(raw_action: Any, valid_actions: list[str], fallback: str) -> str:
if not isinstance(raw_action, str):
return fallback
normalized = raw_action.strip().upper().replace("-", "_").replace(" ", "_")
if normalized in valid_actions:
return normalized
for candidate in valid_actions:
if normalized == candidate.upper():
return candidate
return fallback
def get_llm_action(client: OpenAI, observation) -> tuple[TICEAction, str]:
fallback_b, fallback_t = heuristic_action(observation)
raw_response = ""
try:
completion = client.chat.completions.create(
model=MODEL_NAME,
messages=[
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": build_user_prompt(observation)},
],
temperature=TEMPERATURE,
max_tokens=MAX_TOKENS,
)
raw_response = completion.choices[0].message.content or ""
parsed = json.loads(sanitize_json_response(raw_response))
b_action = coerce_action(
parsed.get("b_cell_action"),
B_CELL_ACTIONS,
fallback_b,
)
t_action = coerce_action(
parsed.get("t_cell_action"),
T_CELL_ACTIONS,
fallback_t,
)
reasoning = str(parsed.get("reasoning", "")).strip() or "no_reasoning"
return TICEAction(b_cell_action=b_action, t_cell_action=t_action), reasoning
except Exception as exc:
fallback_action = TICEAction(
b_cell_action=fallback_b,
t_cell_action=fallback_t,
)
return fallback_action, f"fallback:{type(exc).__name__}"
def require_api_key() -> str:
if API_KEY:
return API_KEY
raise RuntimeError(
"Missing API key. Set HF_TOKEN, API_KEY, or OPENAI_API_KEY before running inference_llm.py."
)
def require_image_name() -> str:
if IMAGE_NAME:
return IMAGE_NAME
raise RuntimeError(
"Missing docker image name. Set LOCAL_IMAGE_NAME (or IMAGE_NAME) before running inference_llm.py."
)
async def run_task(task: str, archetype: str, difficulty: str, client: OpenAI) -> float:
log_start(task=task, env=BENCHMARK, model=MODEL_NAME)
env = await TICEEnv.from_docker_image(require_image_name())
rewards: List[float] = []
steps = 0
score = 0.0
success = False
try:
result = await env.reset(archetype=archetype, difficulty=difficulty)
obs = result.observation
while not obs.done:
action, reasoning = get_llm_action(client, obs)
result = await env.step(action)
obs = result.observation
reward = float(result.reward if result.reward is not None else 0.0)
done = bool(result.done)
steps += 1
rewards.append(reward)
action_summary = (
f"{action.b_cell_action}|{action.t_cell_action}|"
f"phase={obs.episode_phase}|trend={obs.tumor_trend}|note={reasoning[:40]}"
)
log_step(step=steps, action=action_summary, reward=reward, done=done, error=None)
score = (sum(rewards) / len(rewards)) if rewards else 0.0
score = round(float(score), 4)
success = score >= SUCCESS_SCORE_THRESHOLD
except Exception as e:
err = str(e)[:80]
if steps == 0:
log_step(step=1, action="error", reward=0.0, done=True, error=err)
rewards = [0.0]
steps = 1
score = (sum(rewards) / len(rewards)) if rewards else 0.0
score = round(float(score), 4)
success = False
finally:
try:
await env.close()
except Exception:
pass
log_end(success=success, steps=steps, score=score, rewards=rewards)
return float(score)
async def main() -> None:
_ = require_image_name()
llm_client = OpenAI(base_url=API_BASE_URL, api_key=require_api_key())
task_scores: List[float] = []
for task, archetype, difficulty in TASKS:
score = await run_task(
task=task,
archetype=archetype,
difficulty=difficulty,
client=llm_client,
)
task_scores.append(score)
overall = sum(task_scores) / len(task_scores) if task_scores else 0.0
print(f"[DEBUG] overall_score={overall:.3f}", flush=True)
if __name__ == "__main__":
asyncio.run(main())
|