File size: 4,972 Bytes
b214779
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Meta x PyTorch Hackathon - Round 1 Guidelines

## Overview

- **Event**: Meta x PyTorch Hackathon by Scaler School of Technology
- **Theme**: Build OpenEnv environments (Reinforcement Learning)
- **Registration**: 14th March - 3rd April
- **Round 1**: 25th March - 8th April
- **Submission Window Opens**: 28th March
- **Finale**: 25th-26th April
- **Submission Deadline**: 8th April 2026, 11:59 PM (confirm timezone on dashboard)

---

## Team Structure

- **Solo**: Compete individually (locked for Round 1 only)
- **Team**: 2-3 members. Only team lead fills the team form.
- Once confirmed, teams cannot be changed.

---

## Round 1 Problem Statement

Build a complete, real-world OpenEnv environment that an AI agent can learn from through the standard `step()` / `reset()` / `state()` API.

### Key Requirements

1. **Must simulate a real-world task** (not games or toys)
2. **Implement full OpenEnv spec**: typed models, `step()/reset()/state()`, `openenv.yaml`
3. **Minimum 3 tasks** with agent graders (easy → medium → hard, scores/reward 0.0–1.0)
4. **Meaningful reward function** with partial progress signals
5. **Baseline inference script** with reproducible scores
6. **Deploy to Hugging Face Spaces** + working Dockerfile
7. **README** with environment description, action/observation spaces, setup instructions

---

## Evaluation Criteria

### Pre-Submission Checklist (All Must Pass)

| Criteria | Description |
|----------|-------------|
| **HF Space deploys** | Automated ping to Space URL must return 200 and respond to `reset()` |
| **OpenEnv spec compliance** | Validate `openenv.yaml`, typed models, `step()/reset()/state()` endpoints |
| **Dockerfile builds** | Automated docker build on submitted repo |
| **Baseline reproduces** | Run inference script — must complete without error and produce scores |
| **3+ tasks with graders** | Enumerate tasks, run each grader, verify scores/reward in 0.0–1.0 range |

---

## Mandatory Additional Instructions

### Environment Variables (Must be defined)

```bash
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
```

### Inference Script Requirements

- **Filename**: Must be named `inference.py` in the root directory
- **LLM Client**: Must use OpenAI Client for all LLM calls
- **Logging Format**: Must emit structured stdout logs following `[START]`, `[STEP]`, and `[END]` format (field names, ordering, and formatting are strict)

### Infrastructure Restrictions

- Runtime of inference script should be less than **20 minutes**
- Must work on a machine with **vCPU=2, memory=8GB**

---

## Quick Checklist (Must-Haves)

- HF Space returns 200 and responds to `reset()`
- `openenv.yaml` validates; `step()/reset()/state()` endpoints respond correctly
- Dockerfile builds in CI
- `inference.py` runs end-to-end and produces scores
- 3+ tasks with graders; reward in 0.0–1.0 range
- OpenAI client used for all LLM calls; logs follow strict `[START]/[STEP]/[END]` format

---

## Preparatory Course (4 Modules ~3.5 hours)

| Module | Title | Duration |
|--------|-------|----------|
| 1 | Why OpenEnv? | 45 min |
| 2 | Using Existing Environments | 50 min |
| 3 | Deploying Environments | 45 min |
| 4 | Building Your Own Environment | 60 min |

**Note**: Each module - read the README first, then open the notebook in Colab. No local setup needed.

[Course Repository](https://github.com/raun/openenv-course/tree/main)

---

## How to Submit

1. Complete Step 1 (Team/Solo selection)
2. Build your OpenEnv environment
3. Deploy to Hugging Face Spaces
4. Run pre-submission validation script
5. Submit via dashboard (only team leaders can submit)

---

## What Happens After Round 1

- Results announced: 10th April
- Finale: 25th-26th April

---

## Need Help?

- **Email**: help_openenvhackathon@scaler.com
- **Discord**: Join the community for announcements, mentor access, and team matching
- **Discord Link**: https://discord.gg/Dedhy5pkWD

---

## Example Problem Statement Format

> "Build a real-world task environment (e.g., incident triage or logistics scheduling) with clearly defined tasks, automated graders, and reward logic using the OpenEnv framework."

### Expected Deliverables:
- Create an environment an AI agent can interact with
- Define tasks with increasing difficulty
- Write graders that verify task completion
- Define reward logic for scoring
- Package using OpenEnv for automated evaluation

### Evaluation Areas:
- **Runtime correctness**: Runs without errors
- **Interface compliance**: Follows OpenEnv standard
- **Task design**: Clear, realistic, testable
- **Grading logic**: Reward system makes sense

---

## Resources

- [OpenEnv Course Repository](https://github.com/raun/openenv-course/tree/main)
- [Join Discord Community](https://discord.gg/Dedhy5pkWD)
- [Contact Support](mailto:help_openenvhackathon@scaler.com)

---

*Last Updated: April 3, 2026*