PyPatch/main.py at main · Anurag-M1/PyPatch · GitHub

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
import uuid
from typing import Optional

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import HTMLResponse, JSONResponse

from models import (
    PyPatchAction,
    PyPatchObservation,
    ResetRequest,
    ResetResponse,
    StateResponse,
    StepResponse,
    TaskInfo,
)
from tasks import TASKS, TASK_MAP, grade_task

# ─── APP SETUP ────────────────────────────────────────────────────────────────

app = FastAPI(
    title="PyPatch",
    description=(
        "An OpenEnv-compliant RL environment where AI agents fix buggy Python code "
        "across three difficulty levels: easy (syntax), medium (logic), hard (algorithmic)."
    ),
    version="1.0.0",
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

MAX_STEPS_PER_TASK = 5

# ─── SESSION STATE ────────────────────────────────────────────────────────────
# Single-session in-memory state (compatible with single-worker Docker deployment)

_state = {
    "session_id": None,
    "current_task": None,
    "step_count": 0,
    "done": False,
    "best_reward": 0.0,
    "last_reward": 0.0,
}


def _make_observation(task: dict, step_count: int) -> PyPatchObservation:
    return PyPatchObservation(
        task_id=task["id"],
        task_name=task["name"],
        difficulty=task["difficulty"],
        task_description=task["description"],
        buggy_code=task["buggy_code"],
        error_hint=task["error_hint"],
        step_count=step_count,
    )


# ─── ENDPOINTS ────────────────────────────────────────────────────────────────

def _landing_page_html() -> str:
    task_cards = "".join(
        f"""<div class="task-card">
          <div class="task-top">
            <div>
              <h3>{t['name']}</h3>
              <p>{t['description'].splitlines()[0]}</p>
            </div>
            <span class="badge badge-{t['difficulty']}">{t['difficulty']}</span>
          </div>
          <div class="task-meta">
            <span><strong>ID</strong> <code>{t['id']}</code></span>
            <span><strong>Baseline</strong> {t['score']:.2f}</span>
            <span><strong>Grader</strong> <code>{t['grader']}</code></span>
          </div>
        </div>"""
        for t in TASKS
    )
    html = f"""<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
<title>PyPatch 🐛</title>
<style>
  *{{box-sizing:border-box;margin:0;padding:0}}
  body{{font-family:'Segoe UI',system-ui,sans-serif;background:radial-gradient(circle at top,#1a2238 0,#0f1117 45%);color:#e2e8f0;min-height:100vh;padding:2rem;line-height:1.55}}
  a{{color:#7dd3fc;text-decoration:none}}
  a:hover{{text-decoration:underline}}
  .shell{{max-width:1120px;margin:0 auto}}
  .hero{{text-align:center;padding:3.5rem 1rem 2rem}}
  .hero h1{{font-size:3.4rem;font-weight:800;background:linear-gradient(135deg,#fb923c,#ef4444);-webkit-background-clip:text;-webkit-text-fill-color:transparent}}
  .hero p{{color:#94a3b8;font-size:1.1rem;margin-top:.75rem;max-width:780px;margin-left:auto;margin-right:auto}}
  .hero-actions{{display:flex;gap:.8rem;justify-content:center;flex-wrap:wrap;margin-top:1.4rem}}
  .btn{{display:inline-flex;align-items:center;justify-content:center;padding:.8rem 1.1rem;border-radius:10px;font-weight:600;border:1px solid #334155;background:#111827;color:#f8fafc}}
  .btn-primary{{background:linear-gradient(135deg,#f97316,#ef4444);border:none}}
  .badge{{padding:.2rem .6rem;border-radius:999px;font-size:.75rem;font-weight:600;text-transform:uppercase}}
  .badge-easy{{background:#166534;color:#86efac}}
  .badge-medium{{background:#92400e;color:#fde68a}}
  .badge-hard{{background:#7f1d1d;color:#fca5a5}}
  .grid{{display:grid;grid-template-columns:repeat(auto-fit,minmax(220px,1fr));gap:1rem;max-width:1120px;margin:2rem auto}}
  .card,.panel,.task-card{{background:rgba(30,35,48,.88);border:1px solid #2d3748;border-radius:16px;box-shadow:0 12px 30px rgba(0,0,0,.22)}}
  .card{{padding:1.5rem}}
  .card h3{{font-size:.8rem;text-transform:uppercase;color:#64748b;letter-spacing:.1em;margin-bottom:.5rem}}
  .card p{{font-size:1.8rem;font-weight:700;color:#f1f5f9}}
  .card small{{font-size:.85rem;color:#64748b}}
  .section{{max-width:1120px;margin:2.3rem auto}}
  .section h2{{color:#f8fafc;font-size:1.45rem;margin-bottom:.35rem}}
  .section p.lead{{color:#94a3b8;margin-bottom:1rem}}
  .split{{display:grid;grid-template-columns:1.2fr .8fr;gap:1rem}}
  .panel{{padding:1.25rem}}
  .task-list{{display:grid;gap:1rem}}
  .task-card{{padding:1.2rem}}
  .task-top{{display:flex;justify-content:space-between;gap:1rem;align-items:flex-start}}
  .task-card h3{{font-size:1.05rem;margin-bottom:.35rem}}
  .task-card p{{color:#94a3b8;font-size:.95rem}}
  .task-meta{{display:flex;gap:1rem;flex-wrap:wrap;margin-top:1rem;color:#cbd5e1;font-size:.88rem}}
  code{{background:#0f1117;padding:.15rem .4rem;border-radius:4px;font-size:.82rem;color:#f97316}}
  pre{{background:#0b1220;border:1px solid #22304a;border-radius:12px;padding:1rem;overflow:auto;color:#cbd5e1;font-size:.88rem}}
  .endpoints{{max-width:1120px;margin:2rem auto}}
  .endpoints h2{{color:#94a3b8;font-size:.9rem;text-transform:uppercase;letter-spacing:.1em;margin-bottom:1rem}}
  .ep{{display:flex;align-items:center;gap:.75rem;padding:.6rem 1rem;background:#1e2330;border-radius:8px;margin-bottom:.5rem;border:1px solid #2d3748}}
  .method{{font-size:.75rem;font-weight:700;padding:.2rem .5rem;border-radius:4px;min-width:3.5rem;text-align:center}}
  .get{{background:#164e63;color:#7dd3fc}}
  .post{{background:#14532d;color:#86efac}}
  .ep-path{{font-family:monospace;font-size:.9rem;color:#e2e8f0}}
  .ep-desc{{font-size:.8rem;color:#64748b;margin-left:auto}}
  .status-pill{{display:inline-flex;align-items:center;gap:.4rem;background:#14532d;border:1px solid #166534;color:#86efac;padding:.3rem .8rem;border-radius:999px;font-size:.8rem;font-weight:600}}
  .dot{{width:8px;height:8px;background:#22c55e;border-radius:50%;animation:pulse 2s infinite}}
  @keyframes pulse{{0%,100%{{opacity:1}}50%{{opacity:.4}}}}
  .list{{display:grid;gap:.7rem;color:#cbd5e1}}
  .list div{{padding:.75rem .9rem;background:#111827;border:1px solid #243244;border-radius:10px}}
  footer{{text-align:center;color:#475569;font-size:.8rem;margin-top:3rem}}
  @media (max-width: 840px){{.split{{grid-template-columns:1fr}} .hero h1{{font-size:2.6rem}}}}
</style>
</head>
<body>
<div class="shell">
<div class="hero">
  <h1>🐛 PyPatch</h1>
  <p>PyPatch is a judge-friendly RL environment where agents repair buggy Python programs across escalating difficulty levels and receive dense, test-driven reward signals.</p>
  <br/>
  <span class="status-pill"><span class="dot"></span>Live · v1.0.0</span>
  <div class="hero-actions">
    <a class="btn btn-primary" href="/docs">Open API Docs</a>
    <a class="btn" href="/tasks">View Task Metadata</a>
    <a class="btn" href="https://github.com/Anurag-M1/PyPatch">GitHub Repo</a>
  </div>
</div>

<div class="grid">
  <div class="card">
    <h3>Tasks</h3>
    <p>{len(TASKS)}</p>
    <small>easy → medium → hard</small>
  </div>
  <div class="card">
    <h3>Max Steps</h3>
    <p>5</p>
    <small>per episode</small>
  </div>
  <div class="card">
    <h3>Reward</h3>
    <p>0–1</p>
    <small>fraction of tests passed</small>
  </div>
  <div class="card">
    <h3>Baseline</h3>
    <p>0.67</p>
    <small>deterministic average score</small>
  </div>
</div>

<div class="section">
  <h2>Why This Stands Out</h2>
  <p class="lead">Designed for fast evaluation, dense feedback, and clear progression from syntax repair to algorithmic reasoning.</p>
  <div class="split">
    <div class="panel">
      <div class="list">
        <div><strong>Dense reward shaping</strong><br/>Agents receive partial credit instead of only pass/fail outcomes.</div>
        <div><strong>Progressive difficulty</strong><br/>Easy, medium, and hard tasks cover syntax, logic, and algorithmic bug fixing.</div>
        <div><strong>Deterministic grading</strong><br/>Each task uses explicit hidden test cases and reproducible scoring.</div>
        <div><strong>Submission-safe baseline</strong><br/>Baseline inference uses the evaluator proxy and emits structured logs.</div>
      </div>
    </div>
    <div class="panel">
<pre>GET  /        # machine-readable health
GET  /ui     # judge-facing landing page
GET  /tasks  # task metadata + grader + score
POST /reset  # start a task episode
POST /step   # submit fixed code
GET  /state  # inspect current episode</pre>
    </div>
  </div>
</div>

<div class="section">
  <h2>Tasks</h2>
  <p class="lead">Three curated debugging challenges with deterministic graders and non-trivial hidden tests.</p>
  <div class="task-list">{task_cards}</div>
</div>

<div class="section">
  <h2>Judge Demo Flow</h2>
  <p class="lead">A reviewer can understand the project in under two minutes.</p>
  <div class="split">
    <div class="panel">
<pre>1. Open /ui
2. Inspect /tasks for difficulty + grader metadata
3. POST /reset with a task_id
4. POST /step with corrected Python code
5. Observe reward and test breakdown
6. Run inference.py for baseline logs</pre>
    </div>
    <div class="panel">
<pre>curl -X POST http://localhost:7860/reset \\
  -H "Content-Type: application/json" \\
  -d '{{"task_id":"task_medium_second_largest"}}'

curl -X POST http://localhost:7860/step \\
  -H "Content-Type: application/json" \\
  -d '{{"fixed_code":"def second_largest(lst):\\n    unique = sorted(set(lst))\\n    return unique[-2]"}}'</pre>
    </div>
  </div>
</div>

<div class="endpoints">
  <h2>API Endpoints</h2>
  <div class="ep"><span class="method get">GET</span><span class="ep-path">/ui</span><span class="ep-desc">Judge-facing landing page</span></div>
  <div class="ep"><span class="method get">GET</span><span class="ep-path">/health</span><span class="ep-desc">Health check</span></div>
  <div class="ep"><span class="method get">GET</span><span class="ep-path">/tasks</span><span class="ep-desc">List all tasks</span></div>
  <div class="ep"><span class="method post">POST</span><span class="ep-path">/reset</span><span class="ep-desc">Start episode · body: {{"task_id": "..."}}</span></div>
  <div class="ep"><span class="method post">POST</span><span class="ep-path">/step</span><span class="ep-desc">Submit fix · body: {{"fixed_code": "..."}}</span></div>
  <div class="ep"><span class="method get">GET</span><span class="ep-path">/state</span><span class="ep-desc">Current episode state</span></div>
  <div class="ep"><span class="method get">GET</span><span class="ep-path">/docs</span><span class="ep-desc">Swagger UI</span></div>
</div>

<footer>Built by Anurag Singh</footer>
</div>
</body>
</html>"""
    return html


@app.get("/")
async def root():
    """Machine-readable health endpoint used by validators."""
    return {
        "status": "ok",
        "env": "PyPatch",
        "version": app.version,
        "tasks": len(TASKS),
    }


@app.get("/ui", response_class=HTMLResponse)
async def ui():
    """Human-friendly landing page."""
    return HTMLResponse(content=_landing_page_html(), status_code=200)


@app.get("/health")
async def health():
    return {"status": "ok"}


@app.get("/tasks")
async def list_tasks():
    """List all available tasks."""
    return {
        "tasks": [
            TaskInfo(
                id=t["id"],
                name=t["name"],
                difficulty=t["difficulty"],
                description=t["description"],
                grader=t["grader"],
                score=t["score"],
            )
            for t in TASKS
        ]
    }


@app.post("/reset", response_model=ResetResponse)
async def reset(request: Optional[ResetRequest] = None):
    """
    Reset the environment.
    Optionally pass `task_id` to select a specific task.
    Defaults to the easy task.
    """
    task_id = request.task_id if request and request.task_id else TASKS[0]["id"]

    task = TASK_MAP.get(task_id)
    if task is None:
        raise HTTPException(status_code=404, detail=f"Task '{task_id}' not found.")

    _state["session_id"] = str(uuid.uuid4())
    _state["current_task"] = task
    _state["step_count"] = 0
    _state["done"] = False
    _state["best_reward"] = 0.0
    _state["last_reward"] = 0.0

    return ResetResponse(
        observation=_make_observation(task, 0),
        done=False,
        reward=0.0,
    )


@app.post("/step", response_model=StepResponse)
async def step(action: PyPatchAction):
    """
    Submit a fixed version of the buggy code.
    Returns reward (0.0–1.0) based on test cases passed.
    Episode ends when all tests pass OR max steps reached.
    """
    if _state["current_task"] is None:
        raise HTTPException(status_code=400, detail="Environment not initialised. Call /reset first.")

    if _state["done"]:
        raise HTTPException(status_code=400, detail="Episode complete. Call /reset to start a new episode.")

    _state["step_count"] += 1
    task = _state["current_task"]

    reward, info = grade_task(task, action.fixed_code)
    done = reward >= 1.0 or _state["step_count"] >= MAX_STEPS_PER_TASK

    _state["best_reward"] = max(_state["best_reward"], reward)
    _state["last_reward"] = reward
    _state["done"] = done

    return StepResponse(
        observation=_make_observation(task, _state["step_count"]),
        reward=reward,
        done=done,
        info={
            "tests_passed": info["passed"],
            "tests_total": info["total"],
            "test_results": info["results"],
            "exec_error": info.get("error"),
            "steps_remaining": max(0, MAX_STEPS_PER_TASK - _state["step_count"]),
        },
    )


@app.get("/state", response_model=StateResponse)
async def get_state():
    """Return current environment state."""
    return StateResponse(
        task_id=_state["current_task"]["id"] if _state["current_task"] else None,
        step_count=_state["step_count"],
        done=_state["done"],
        total_reward=_state["best_reward"],
        session_id=_state["session_id"],
    )