File size: 7,799 Bytes
57fc36e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import argparse
import json
import os
import random
from dataclasses import dataclass
from typing import Any

import requests
from openai import OpenAI


TASKS = ["easy_docker", "medium_k8s", "hard_ml_config"]


@dataclass
class EpisodeResult:
    task_id: str
    final_score: float
    done: bool
    steps: int
    rewards: list[float]


def build_openai_client() -> OpenAI:
    api_key = os.getenv("OPENAI_API_KEY")
    if not api_key:
        raise RuntimeError("OPENAI_API_KEY is required for OpenAI baseline mode")
    return OpenAI(api_key=api_key)


def extract_json_object(text: str) -> dict[str, Any]:
    text = text.strip()
    if "```" in text:
        blocks = text.split("```")
        for block in blocks:
            block = block.strip()
            if block.startswith("json"):
                block = block[4:].strip()
            if block.startswith("{") and block.endswith("}"):
                return json.loads(block)
    start = text.find("{")
    end = text.rfind("}")
    if start != -1 and end != -1 and end > start:
        return json.loads(text[start : end + 1])
    raise ValueError("No JSON object found in model output")


def choose_heuristic_action(task_id: str, step: int) -> dict[str, Any]:
    # Deterministic policy for reproducible baseline.
    easy_plan = [
        {"operation": "edit", "path": "services.web.image", "value": "nginx:latest"},
        {"operation": "delete", "path": "services.web.ports.1"},
        {"operation": "edit", "path": "services.web.environment", "value": {"DEBUG": "true", "API_KEY": "placeholder"}},
        {"operation": "edit", "path": "services.db.ports.0", "value": "5432:5432"},
    ]

    medium_plan = [
        {"operation": "edit", "path": "metadata.namespace", "value": "default"},
        {"operation": "edit", "path": "spec.replicas", "value": 3},
        {"operation": "edit", "path": "spec.template.spec.containers.0.image", "value": "nginx:latest"},
        {"operation": "edit", "path": "spec.template.spec.containers.0.resources.limits.memory", "value": "512Mi"},
        {"operation": "edit", "path": "spec.template.spec.containers.0.resources.requests.memory", "value": "256Mi"},
        {"operation": "edit", "path": "spec.template.spec.containers.0.resources.requests.cpu", "value": "500m"},
        {"operation": "add", "path": "spec.template.spec.containers.0.ports", "value": [{"containerPort": 80}]},
    ]

    hard_plan = [
        {"operation": "delete", "path": "training.fp16"},
        {"operation": "edit", "path": "training.batch_size", "value": 16},
        {"operation": "edit", "path": "training.gradient_accumulation_steps", "value": 2},
        {"operation": "edit", "path": "training.max_steps", "value": 1000},
        {"operation": "edit", "path": "training.warmup_steps", "value": 100},
        {"operation": "edit", "path": "training.optimizer.type", "value": "adamw"},
        {"operation": "edit", "path": "hardware.gpu_count", "value": 1},
        {"operation": "edit", "path": "data.train_batch_size", "value": 32},
        {"operation": "edit", "path": "logging.log_interval", "value": 10},
    ]

    plans = {
        "easy_docker": easy_plan,
        "medium_k8s": medium_plan,
        "hard_ml_config": hard_plan,
    }
    plan = plans[task_id]
    return plan[min(step, len(plan) - 1)]


def choose_openai_action(client: OpenAI, model: str, observation: dict[str, Any]) -> dict[str, Any]:
    system_prompt = (
        "You are an environment-control agent for configuration debugging. "
        "Return exactly one JSON object action."
    )
    user_prompt = (
        "Task:\n"
        f"{observation['task_description']}\n\n"
        "Allowed schema:\n"
        "{\"operation\": \"edit|add|delete\", \"path\": \"dot.path\", \"value\": any|null}\n\n"
        f"Current score: {observation['overall_score']}\n"
        f"Validation errors: {observation['validation_errors']}\n"
        f"Current YAML:\n{observation['current_config']}\n"
    )

    response = client.chat.completions.create(
        model=model,
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt},
        ],
        temperature=0,
        top_p=1,
        seed=42,
    )
    content = response.choices[0].message.content or ""
    return extract_json_object(content)


def run_episode(
    api_base_url: str,
    task_id: str,
    max_steps: int,
    policy: str,
    model: str,
    openai_client: OpenAI | None,
) -> EpisodeResult:
    reset_resp = requests.post(f"{api_base_url}/reset", json={"task_id": task_id}, timeout=30)
    reset_resp.raise_for_status()
    observation = reset_resp.json()["observation"]

    rewards: list[float] = []
    done = False

    print(f"[START] task={task_id} policy={policy}")

    for step in range(max_steps):
        if done:
            break

        if policy == "heuristic":
            action = choose_heuristic_action(task_id, step)
        else:
            assert openai_client is not None
            action = choose_openai_action(openai_client, model, observation)

        step_resp = requests.post(f"{api_base_url}/step", json=action, timeout=30)
        if step_resp.status_code != 200:
            rewards.append(0.0)
            print(f"[STEP] task={task_id} step={step} action=invalid reward=0.00 done=false")
            continue

        payload = step_resp.json()
        observation = payload["observation"]
        reward = payload["reward"]
        done = payload["done"]
        reward_value = float(reward["value"])
        rewards.append(reward_value)

        print(
            f"[STEP] task={task_id} step={step} action={action.get('operation')}:{action.get('path')} "
            f"reward={reward_value:.3f} score={observation['overall_score']:.3f} done={str(done).lower()}"
        )

    result = EpisodeResult(
        task_id=task_id,
        final_score=float(observation["overall_score"]),
        done=done,
        steps=min(max_steps, len(rewards)),
        rewards=rewards,
    )

    reward_text = ",".join(f"{v:.3f}" for v in rewards)
    print(
        f"[END] task={task_id} score={result.final_score:.3f} "
        f"steps={result.steps} done={str(result.done).lower()} rewards={reward_text}"
    )
    return result


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description="Baseline inference for ConfigDebuggerEnv")
    parser.add_argument("--api-base-url", default=os.getenv("API_BASE_URL", "http://localhost:8000"))
    parser.add_argument("--max-steps", type=int, default=12)
    parser.add_argument("--policy", choices=["heuristic", "openai"], default="heuristic")
    parser.add_argument("--model", default=os.getenv("OPENAI_MODEL", "gpt-4o-mini"))
    parser.add_argument("--seed", type=int, default=42)
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    random.seed(args.seed)

    openai_client: OpenAI | None = None
    if args.policy == "openai":
        openai_client = build_openai_client()

    results: list[EpisodeResult] = []
    for task_id in TASKS:
        results.append(
            run_episode(
                api_base_url=args.api_base_url,
                task_id=task_id,
                max_steps=args.max_steps,
                policy=args.policy,
                model=args.model,
                openai_client=openai_client,
            )
        )

    avg = sum(r.final_score for r in results) / len(results)
    print("\n=== BASELINE SUMMARY ===")
    for result in results:
        print(
            f"{result.task_id}: final_score={result.final_score:.3f} steps={result.steps} done={str(result.done).lower()}"
        )
    print(f"average_score={avg:.3f}")


if __name__ == "__main__":
    main()