File size: 11,484 Bytes
8b4d6a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
186ab8c
 
 
 
 
 
8b4d6a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Inference Script β€” Annotation QA Environment
=============================================
MANDATORY
- Before submitting, ensure the following variables are defined:
    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.

- Defaults are set only for API_BASE_URL and MODEL_NAME:
    API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")
    MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-72B-Instruct")

- The inference script must be named `inference.py` and placed in the root
- Participants must use OpenAI Client for all LLM calls
- Participants must emit structured stdout logs strictly following [START],
  [STEP], and [END] format

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>
"""

import asyncio
import json
import os
import sys
import textwrap
from typing import Any, Dict, List, Optional

from openai import OpenAI

# Add parent to path for imports
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
try:
    from annotation_qa_env.models import AnnotationQAAction, AnnotationQAObservation
    from annotation_qa_env.server.environment import AnnotationQAEnvironment
except ImportError:
    from models import AnnotationQAAction, AnnotationQAObservation
    from server.environment import AnnotationQAEnvironment

# ──────────────────────────────────────────────
# Configuration
# ──────────────────────────────────────────────

LOCAL_IMAGE_NAME = os.getenv("LOCAL_IMAGE_NAME")
HF_TOKEN = os.getenv("HF_TOKEN")
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 = "annotation_qa_env"
TASKS = ["fix_bboxes", "fix_classes", "batch_audit"]
MAX_STEPS_PER_TASK = {"fix_bboxes": 15, "fix_classes": 20, "batch_audit": 30}
TEMPERATURE = 0.3
MAX_TOKENS = 500
SUCCESS_SCORE_THRESHOLD = 0.1

SYSTEM_PROMPT = textwrap.dedent("""
You are an AI annotation quality reviewer. You examine synthetic scene
annotations and fix errors in bounding boxes and class labels.

You will receive:
1. A scene description with objects and their true positions
2. Current annotations (some may have errors)
3. Available classes

Your job: Compare annotations against the scene description and fix errors.

AVAILABLE ACTIONS (respond with valid JSON):
- {"action_type": "adjust_bbox", "annotation_id": <id>, "new_bbox": [x, y, w, h]}
- {"action_type": "change_class", "annotation_id": <id>, "new_class": "<class>"}
- {"action_type": "add_annotation", "new_bbox": [x, y, w, h], "new_class": "<class>"}
- {"action_type": "remove_annotation", "annotation_id": <id>}
- {"action_type": "submit"}

All bbox values are normalized to 0.0–1.0.

STRATEGY:
1. Compare each annotation's bbox against the scene objects' bboxes
2. Check if class labels match the scene objects
3. Look for spurious annotations that don't match any scene object
4. Look for scene objects that have no annotation
5. Fix errors one at a time, then submit

RESPOND WITH ONLY A SINGLE JSON ACTION, no explanation.
""").strip()


# ──────────────────────────────────────────────
# Logging helpers (exact format from problem statement)
# ──────────────────────────────────────────────

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:.2f} rewards={rewards_str}",
        flush=True,
    )


# ──────────────────────────────────────────────
# Prompt building
# ──────────────────────────────────────────────

def build_user_prompt(obs: AnnotationQAObservation) -> str:
    """Build the user prompt from the observation."""
    # Format scene objects
    scene_desc = obs.scene_description

    # Format current annotations
    ann_lines = []
    for ann in obs.annotations:
        ann_lines.append(
            f"  ID={ann.id}: class='{ann.class_label}', "
            f"bbox=[{ann.bbox[0]:.3f}, {ann.bbox[1]:.3f}, {ann.bbox[2]:.3f}, {ann.bbox[3]:.3f}]"
        )
    annotations_str = "\n".join(ann_lines) if ann_lines else "  (none)"

    # Format scene ground truth objects
    obj_lines = []
    for obj in obs.scene_objects:
        bbox = obj.get("bbox", [0, 0, 0, 0])
        obj_lines.append(
            f"  {obj['class_label']} at {obj.get('position', '?')}: "
            f"bbox=[{bbox[0]:.3f}, {bbox[1]:.3f}, {bbox[2]:.3f}, {bbox[3]:.3f}]"
        )
    objects_str = "\n".join(obj_lines) if obj_lines else "  (none)"

    prompt = f"""Task: {obs.task_description}
Step {obs.step_count}/{obs.max_steps} | Corrections made: {obs.corrections_made}
Feedback: {obs.message}

SCENE OBJECTS (ground truth):
{objects_str}

CURRENT ANNOTATIONS (may have errors):
{annotations_str}

AVAILABLE CLASSES: {', '.join(obs.available_classes)}

Compare annotations against scene objects. Find and fix ONE error, or submit if all are correct.
Respond with a single JSON action."""

    return prompt


def parse_llm_response(response_text: str) -> AnnotationQAAction:
    """Parse the LLM's JSON response into an action."""
    # Try to extract JSON from the response
    text = response_text.strip()

    # Handle common LLM formatting issues
    if text.startswith("```json"):
        text = text[7:]
    if text.startswith("```"):
        text = text[3:]
    if text.endswith("```"):
        text = text[:-3]
    text = text.strip()

    try:
        data = json.loads(text)
    except json.JSONDecodeError:
        # Try to find JSON in the response
        import re
        json_match = re.search(r'\{[^}]+\}', text)
        if json_match:
            try:
                data = json.loads(json_match.group())
            except json.JSONDecodeError:
                # Fallback: submit
                return AnnotationQAAction(action_type="submit")
        else:
            return AnnotationQAAction(action_type="submit")

    return AnnotationQAAction(
        action_type=data.get("action_type", "submit"),
        annotation_id=data.get("annotation_id"),
        new_bbox=data.get("new_bbox"),
        new_class=data.get("new_class"),
    )


# ──────────────────────────────────────────────
# LLM interaction
# ──────────────────────────────────────────────

def get_model_action(
    client: OpenAI,
    obs: AnnotationQAObservation,
) -> AnnotationQAAction:
    """Query the LLM for the next action."""
    user_prompt = build_user_prompt(obs)

    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()
        return parse_llm_response(text)
    except Exception as exc:
        print(f"[DEBUG] Model request failed: {exc}", flush=True)
        return AnnotationQAAction(action_type="submit")


# ──────────────────────────────────────────────
# Main loop
# ──────────────────────────────────────────────

def run_task(client: OpenAI, env: AnnotationQAEnvironment, task_name: str) -> float:
    """Run a single task and return the score."""
    max_steps = MAX_STEPS_PER_TASK.get(task_name, 20)
    rewards: List[float] = []
    steps_taken = 0
    score = 0.0
    success = False

    log_start(task=task_name, env=BENCHMARK, model=MODEL_NAME)

    try:
        # Reset environment with the specific task
        obs = env.reset(task=task_name, seed=42)
        last_reward = 0.0

        for step in range(1, max_steps + 1):
            if obs.done:
                break

            # Get action from LLM
            action = get_model_action(client, obs)
            action_str = f"{action.action_type}"
            if action.annotation_id is not None:
                action_str += f"(id={action.annotation_id})"

            # Execute action
            obs = env.step(action)

            reward = obs.reward if obs.reward is not None else 0.0
            done = obs.done
            error = obs.last_action_error

            rewards.append(reward)
            steps_taken = step
            last_reward = reward

            log_step(
                step=step,
                action=action_str,
                reward=reward,
                done=done,
                error=error,
            )

            if done:
                break

        # Compute final score: use the last reward (which is the grader score on submit/timeout)
        if rewards:
            score = rewards[-1]  # Last reward is the final grade
        score = max(0.0, min(1.0, score))
        success = score >= SUCCESS_SCORE_THRESHOLD

    except Exception as exc:
        print(f"[DEBUG] Task {task_name} error: {exc}", flush=True)
        score = 0.0
        success = False

    log_end(success=success, steps=steps_taken, score=score, rewards=rewards)
    return score


def main() -> None:
    """Run inference on all 3 tasks."""
    client = OpenAI(base_url=API_BASE_URL, api_key=HF_TOKEN)
    env = AnnotationQAEnvironment()

    total_score = 0.0
    for task_name in TASKS:
        print(f"\n{'='*60}", flush=True)
        print(f"Running task: {task_name}", flush=True)
        print(f"{'='*60}", flush=True)
        score = run_task(client, env, task_name)
        total_score += score
        print(f"Task {task_name} score: {score:.3f}\n", flush=True)

    avg_score = total_score / len(TASKS)
    print(f"\n{'='*60}", flush=True)
    print(f"Average score across {len(TASKS)} tasks: {avg_score:.3f}", flush=True)
    print(f"{'='*60}", flush=True)


if __name__ == "__main__":
    main()