anonajay commited on
Commit
3a2da6a
·
verified ·
1 Parent(s): 80595a5

Upload folder using huggingface_hub

Browse files
inference.py CHANGED
@@ -50,13 +50,12 @@ from typing import List, Optional
50
  from openai import OpenAI
51
  from dotenv import load_dotenv
52
 
53
- # 🚀 THE FIX: Import the base EnvClient. It is async by default.
54
- from openenv import EnvClient
 
55
 
56
- # Load local .env for testing
57
  load_dotenv()
58
 
59
- # IMPORT YOUR ACTION CLASS
60
  try:
61
  from models import LintCodingAgentAction
62
  except ImportError:
@@ -72,7 +71,7 @@ TASK_NAME = "multi-lang-lint-v1"
72
  BENCHMARK = "lint-coding-v1"
73
  MAX_STEPS = 10
74
 
75
- SYSTEM_PROMPT = "You are an expert architect. Fix the code in the VFS. Return ONLY raw code."
76
 
77
  def log_start(task: str, env: str, model: str) -> None:
78
  print(f"[START] task={task} env={env} model={model}", flush=True)
@@ -83,11 +82,22 @@ def log_step(step: int, action: str, reward: float, done: bool, error: Optional[
83
 
84
  def log_end(success: bool, steps: int, score: float, rewards: List[float]) -> None:
85
  rewards_str = ",".join(f"{r:.2f}" for r in rewards)
86
- print(f"[END] success={str(success).lower()} steps={steps} score={score:.3f} rewards={rewards_str}", flush=True)
87
 
88
- async def main() -> None:
89
  client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
90
 
 
 
 
 
 
 
 
 
 
 
 
91
  rewards: List[float] = []
92
  steps_taken = 0
93
  success = False
@@ -95,60 +105,54 @@ async def main() -> None:
95
  log_start(task=TASK_NAME, env=BENCHMARK, model=MODEL_NAME)
96
 
97
  try:
98
- # 🚀 THE CRITICAL HANDSHAKE:
99
- # Use EnvClient directly. Use 'base_url' as the keyword.
100
- async with EnvClient(base_url=ADDRESS) as env:
 
 
 
 
101
 
102
- # Reset Environment
103
- result = await env.reset()
104
-
105
- for step in range(1, MAX_STEPS + 1):
106
- obs = result.observation
107
-
108
- try:
109
- vfs_display = json.dumps(json.loads(obs.code_context), indent=2)
110
- except:
111
- vfs_display = obs.code_context
112
-
113
- user_prompt = f"VFS Structure:\n{vfs_display}\n\nTask: {obs.problem_statement}"
114
-
115
- # LLM Call
116
- completion = client.chat.completions.create(
117
- model=MODEL_NAME,
118
- messages=[
119
- {"role": "system", "content": SYSTEM_PROMPT},
120
- {"role": "user", "content": user_prompt},
121
- ],
122
- temperature=0.2,
123
- max_tokens=200,
124
- )
125
-
126
- action_text = (completion.choices[0].message.content or "").strip()
127
-
128
- # State Transition (Awaited)
129
- result = await env.step(LintCodingAgentAction(
130
- code_solution=action_text,
131
- explanation=f"Async fix level {obs.level}"
132
- ))
133
-
134
- reward = result.reward or 0.0
135
- done = result.done
136
-
137
- rewards.append(reward)
138
- steps_taken = step
139
- log_step(step, action_text, reward, done, None)
140
-
141
- if done:
142
- success = True
143
- break
144
 
145
  except Exception as e:
146
- if steps_taken == 0:
147
- log_step(1, "error_retry", 0.0, True, str(e))
148
 
149
  finally:
150
- score = sum(rewards) / len(rewards) if rewards else 0.0
 
151
  log_end(success, steps_taken, score, rewards)
152
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
153
  if __name__ == "__main__":
154
- asyncio.run(main())
 
50
  from openai import OpenAI
51
  from dotenv import load_dotenv
52
 
53
+ # 🚀 THE BULLETPROOF IMPORT
54
+ import openenv
55
+ from openenv import SyncEnvClient
56
 
 
57
  load_dotenv()
58
 
 
59
  try:
60
  from models import LintCodingAgentAction
61
  except ImportError:
 
71
  BENCHMARK = "lint-coding-v1"
72
  MAX_STEPS = 10
73
 
74
+ SYSTEM_PROMPT = "You are a lead architect. Fix the code in the VFS. Return raw code only."
75
 
76
  def log_start(task: str, env: str, model: str) -> None:
77
  print(f"[START] task={task} env={env} model={model}", flush=True)
 
82
 
83
  def log_end(success: bool, steps: int, score: float, rewards: List[float]) -> None:
84
  rewards_str = ",".join(f"{r:.2f}" for r in rewards)
85
+ print(f"[END] success={str(success).lower()} steps={steps} score={score:.2f} rewards={rewards_str}", flush=True)
86
 
87
+ def main() -> None:
88
  client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
89
 
90
+ # 🚀 KEYWORD AGNOSTIC INITIALIZATION
91
+ # We try positional first, then 'address', then 'base_url'
92
+ env = None
93
+ try:
94
+ env = SyncEnvClient(ADDRESS)
95
+ except TypeError:
96
+ try:
97
+ env = SyncEnvClient(address=ADDRESS)
98
+ except TypeError:
99
+ env = SyncEnvClient(base_url=ADDRESS)
100
+
101
  rewards: List[float] = []
102
  steps_taken = 0
103
  success = False
 
105
  log_start(task=TASK_NAME, env=BENCHMARK, model=MODEL_NAME)
106
 
107
  try:
108
+ # Check if the environment requires a context manager
109
+ # If .sync() exists, we use it; otherwise, we call reset directly.
110
+ if hasattr(env, 'sync'):
111
+ with env.sync() as synced_env:
112
+ execute_loop(synced_env, client, rewards)
113
+ else:
114
+ execute_loop(env, client, rewards)
115
 
116
+ success = True if rewards and any(r > 0 for r in rewards) else False
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
117
 
118
  except Exception as e:
119
+ log_step(1, "system_error", 0.0, True, str(e))
 
120
 
121
  finally:
122
+ steps_taken = len(rewards)
123
+ score = sum(rewards) / steps_taken if steps_taken > 0 else 0.0
124
  log_end(success, steps_taken, score, rewards)
125
 
126
+ def execute_loop(env, client, rewards):
127
+ """The actual logic loop shared between sync/synced-context modes"""
128
+ result = env.reset()
129
+ for step in range(1, MAX_STEPS + 1):
130
+ obs = result.observation
131
+ vfs_display = obs.code_context # Already JSON stringified in our server
132
+
133
+ user_prompt = f"VFS:\n{vfs_display}\n\nTask: {obs.problem_statement}"
134
+
135
+ completion = client.chat.completions.create(
136
+ model=MODEL_NAME,
137
+ messages=[
138
+ {"role": "system", "content": SYSTEM_PROMPT},
139
+ {"role": "user", "content": user_prompt},
140
+ ],
141
+ temperature=0.2,
142
+ )
143
+
144
+ action_text = completion.choices[0].message.content.strip()
145
+
146
+ result = env.step(LintCodingAgentAction(
147
+ code_solution=action_text,
148
+ explanation=f"Architect fix"
149
+ ))
150
+
151
+ rewards.append(result.reward or 0.0)
152
+ log_step(step, action_text, result.reward, result.done, None)
153
+
154
+ if result.done:
155
+ break
156
+
157
  if __name__ == "__main__":
158
+ main()
server/templates/level_1/core.py ADDED
File without changes
server/templates/level_1/main.py ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ # BROKEN: This will fail if run as a module without the dot
2
+ import core
3
+
4
+ def run():
5
+ print("Starting...")
server/templates/level_10/middleware.py ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import asyncio
2
+
3
+ # BROKEN: This global variable is shared across all async tasks.
4
+ # Task A will set this, then 'await'. Task B will then overwrite it.
5
+ request_store = {"id": None}
6
+
7
+ async def process_request(req_id):
8
+ request_store["id"] = req_id
9
+
10
+ # Simulate a database call or API hit
11
+ await asyncio.sleep(1)
12
+
13
+ # By the time we get here, another request might have
14
+ # changed request_store["id"]
15
+ return f"Finished request: {request_store['id']}"
16
+
17
+ async def main():
18
+ # Simulate two concurrent requests
19
+ results = await asyncio.gather(
20
+ process_request("REQ-100"),
21
+ process_request("REQ-200")
22
+ )
23
+ print(results) # Likely to show ['Finished request: REQ-200', 'Finished request: REQ-200']
server/templates/level_11/handler.py ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ # The needle in the haystack
2
+ def verify_token(token):
3
+ # This is the logic the Master Agent needs to find
4
+ import os
5
+ secret = os.getenv("JWT_SECRET")
6
+ return token == secret
server/templates/level_11/orchestrator.py ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ # MASTER AGENT STARTING POINT
2
+ # Task: Find the verification logic and document the Auth Flow.
3
+ # Error: Repo size exceeds single-pass context.
4
+ import os
5
+
6
+ def start_audit():
7
+ print("Initiating repository-wide security audit...")
8
+ # The agent needs to trigger a 'spawn_sub_agent' call here
server/templates/level_12/deployer.py ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # BROKEN: This script currently deploys without checking conditions.
2
+ # High-level architecture requirement:
3
+ # If storm_risk > 80%, abort to prevent data center power instability.
4
+
5
+ def run_deployment():
6
+ print("Starting production deployment...")
7
+ # TODO: Check Weather MCP here
8
+ print("Deployment Successful!")
9
+
10
+ if __name__ == "__main__":
11
+ run_deployment()
server/templates/level_12/mcp_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tools": [
3
+ {
4
+ "name": "get_weather",
5
+ "parameters": {"location": "string"},
6
+ "description": "Returns current weather metrics including storm_risk (0-100)."
7
+ }
8
+ ]
9
+ }
server/templates/level_13/ai_client.py ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ import openai
2
+
3
+ # BROKEN: This is the old v0.28 syntax.
4
+ # It will throw an AttributeError in the new v1.0+ environment.
5
+ def get_ai_response(prompt):
6
+ response = openai.ChatCompletion.create(
7
+ model="gpt-3.5-turbo",
8
+ messages=[{"role": "user", "content": prompt}]
9
+ )
10
+ return response.choices[0].message.content
server/templates/level_13/requirenments.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ openai==1.3.0 # Note the version: this forces the break
server/templates/level_14/auth.py ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ def check_user_access(user):
2
+ <<<<<<< HEAD
3
+ # Current Version: Fast but less secure
4
+ return user.is_authenticated
5
+ =======
6
+ # Incoming Version: Secure version using MFA check
7
+ if not user.is_authenticated:
8
+ return False
9
+ return user.has_mfa_enabled()
10
+ >>>>>>> feature/mfa-upgrade
server/templates/level_14/session.py ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ class User:
2
+ def __init__(self, auth_status, mfa_status):
3
+ self.is_authenticated = auth_status
4
+ self.mfa_status = mfa_status
5
+
6
+ def has_mfa_enabled(self):
7
+ return self.mfa_status
server/templates/level_15/backend_config.yaml ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ # BROKEN: The SUPABASE_URL is missing, which causes the API deployment tool to crash.
2
+ service: backend_api
3
+ runtime: nodejs20
4
+ env_vars:
5
+ - name: PORT
6
+ value: 8080
7
+ - name: SUPABASE_URL
8
+ value: "" # <--- AGENT MUST FIX THIS
server/templates/level_15/deploynment_log.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ [SYSTEM] Database deployed at: https://xyz-db.supabase.co
2
+ [ERROR] backend_api failed to start: Missing SUPABASE_URL.
server/templates/level_15/infra_manifest.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "services": ["database", "backend_api", "frontend_ui"],
3
+ "dependencies": {
4
+ "backend_api": ["database"],
5
+ "frontend_ui": ["backend_api"]
6
+ }
7
+ }
server/templates/level_2/main.js ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ // BROKEN: You can't 'require' an ES Module
2
+ const { log } = require('./utils.js');
3
+ log("Hello World");
server/templates/level_2/utils.js ADDED
@@ -0,0 +1 @@
 
 
1
+ export const log = (msg) => console.log(msg);
server/templates/level_3/processor.py ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ # BROKEN: The default list [] is created once at definition time.
2
+ def add_to_box(item, box=[]):
3
+ box.append(item)
4
+ return box
5
+
6
+ def clear_and_process(data):
7
+ # This should return a fresh list every time,
8
+ # but currently it accumulates data from previous calls.
9
+ return add_to_box(data)
server/templates/level_3/test_logic.py ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from processor import add_to_box
2
+
3
+ def test_accumulation():
4
+ print("Running Test 1...")
5
+ res1 = add_to_box("A")
6
+ print(f"Call 1: {res1}")
7
+
8
+ print("Running Test 2...")
9
+ res2 = add_to_box("B")
10
+ print(f"Call 2: {res2}")
11
+
12
+ # If the bug is fixed, res2 should only be ['B']
13
+ if len(res2) > 1:
14
+ print("FAIL: List is accumulating items!")
15
+ return False
16
+ return True
17
+
18
+ if __name__ == "__main__":
19
+ test_accumulation()
server/templates/level_4/main.cpp ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ #include "shapes.h"
2
+ #include "utils.h" // This indirectly includes shapes.h again!
3
+
4
+ int main() {
5
+ Circle my_circle = {5.0};
6
+ return 0;
7
+ }
server/templates/level_4/shapes.h ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ // BROKEN: No header guards here
2
+ struct Circle {
3
+ float radius;
4
+ };
5
+
6
+ float calculate_area(Circle c);
server/templates/level_4/utils.h ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ #include "shapes.h"
2
+
3
+ void print_shape_info(Circle c);
server/templates/level_5/com/security/AuthService.java ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ package com.security;
2
+
3
+ public class AuthService {
4
+ // BROKEN: 'private' prevents the test suite from seeing this method
5
+ private boolean validateToken(String token) {
6
+ return token.equals("lint-ai-2026");
7
+ }
8
+ }
server/templates/level_5/com/tests/SecurityTest.java ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ package com.tests;
2
+ import com.security.AuthService;
3
+
4
+ public class SecurityTest {
5
+ public void testLogin() {
6
+ AuthService auth = new AuthService();
7
+ // This line causes a "method has private access" compiler error
8
+ if (auth.validateToken("lint-ai-2026")) {
9
+ System.out.println("Test Passed");
10
+ }
11
+ }
12
+ }
server/templates/level_6/models.py ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ from django.db import models
2
+
3
+ class Author(models.Model):
4
+ name = models.CharField(max_length=100)
5
+
6
+ class Post(models.Model):
7
+ title = models.CharField(max_length=100)
8
+ author = models.ForeignKey(Author, on_delete=models.CASCADE)
server/templates/level_6/view.py ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from .models import Post
2
+
3
+ def get_post_data():
4
+ # BROKEN: This is the N+1 trigger.
5
+ # It fetches posts, but not the related authors.
6
+ posts = Post.objects.all()
7
+
8
+ results = []
9
+ for post in posts:
10
+ # Each access of 'post.author' triggers a NEW database hit!
11
+ results.append({
12
+ 'title': post.title,
13
+ 'author_name': post.author.name
14
+ })
15
+ return results
server/templates/level_7/main.rs ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ use std::thread;
2
+
3
+ fn main() {
4
+ let mut counter = 0;
5
+ let mut handles = vec![];
6
+
7
+ for _ in 0..10 {
8
+ // BROKEN: 'counter' cannot be moved into multiple threads.
9
+ // Rust's borrow checker will stop this from compiling.
10
+ let handle = thread::spawn(move || {
11
+ counter += 1;
12
+ });
13
+ handles.push(handle);
14
+ }
15
+
16
+ for handle in handles {
17
+ handle.join().unwrap();
18
+ }
19
+
20
+ println!("Result: {}", counter);
21
+ }
server/templates/level_8/InfiniteScroll.js ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import React, { useEffect, useRef } from 'react';
2
+
3
+ const InfiniteScroll = ({ loadMore }) => {
4
+ const loaderRef = useRef(null);
5
+
6
+ useEffect(() => {
7
+ const observer = new IntersectionObserver((entries) => {
8
+ if (entries[0].isIntersecting) {
9
+ loadMore();
10
+ }
11
+ });
12
+
13
+ if (loaderRef.current) {
14
+ observer.observe(loaderRef.current);
15
+ }
16
+
17
+ // BROKEN: Missing cleanup function!
18
+ // The observer will keep watching the DOM even if
19
+ // this component is destroyed.
20
+ }, [loadMore]);
21
+
22
+ return <div ref={loaderRef}>Loading more posts...</div>;
23
+ };
24
+
25
+ export default InfiniteScroll;
server/templates/level_9/processor.go ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ package main
2
+
3
+ import "fmt"
4
+
5
+ func main() {
6
+ // BROKEN: An unbuffered channel requires a receiver to be ready
7
+ // at the exact moment a sender sends.
8
+ ch := make(chan int)
9
+
10
+ ch <- 42 // Execution stops here forever!
11
+
12
+ val := <-ch
13
+ fmt.Println("Received:", val)
14
+ }