Spaces:
Sleeping
Sleeping
fix: remove duplicate YAML frontmatter in README
Browse files
README.md
CHANGED
|
@@ -10,14 +10,28 @@ tags:
|
|
| 10 |
- sql
|
| 11 |
- reinforcement-learning
|
| 12 |
- contextual-bandit
|
|
|
|
|
|
|
| 13 |
---
|
| 14 |
-
---
|
| 15 |
-
title: Sql Agent Openenv
|
| 16 |
-
emoji: 🏢
|
| 17 |
-
colorFrom: yellow
|
| 18 |
-
colorTo: green
|
| 19 |
-
sdk: docker
|
| 20 |
-
pinned: false
|
| 21 |
-
---
|
| 22 |
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
- sql
|
| 11 |
- reinforcement-learning
|
| 12 |
- contextual-bandit
|
| 13 |
+
- linucb
|
| 14 |
+
- gepa
|
| 15 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
+
# SQL Agent OpenEnv
|
| 18 |
+
|
| 19 |
+
An OpenEnv-compliant RL environment for SQL generation, featuring:
|
| 20 |
+
|
| 21 |
+
- **LinUCB contextual bandit** — selects repair strategies based on error context
|
| 22 |
+
- **GEPA (Generative Evolutionary Prompt Adaptation)** — evolves the system prompt from failure patterns
|
| 23 |
+
- **3 difficulty tiers** — easy / medium / hard benchmark tasks on a SQLite e-commerce schema
|
| 24 |
+
- **Shaped reward function** — success bonus, attempt penalty, error severity signal
|
| 25 |
+
- **HER (Hindsight Experience Replay)** — relabels failed episodes for richer signal
|
| 26 |
+
|
| 27 |
+
## OpenEnv API
|
| 28 |
+
|
| 29 |
+
| Endpoint | Method | Description |
|
| 30 |
+
|---|---|---|
|
| 31 |
+
| `/reset` | POST | Start a new episode |
|
| 32 |
+
| `/step` | POST | Execute one repair action |
|
| 33 |
+
| `/state` | GET | Current environment state |
|
| 34 |
+
|
| 35 |
+
## Demo
|
| 36 |
+
|
| 37 |
+
Click **Demo** in the top-right to watch the agent fail, self-repair via RL, then improve through two GEPA prompt-evolution cycles (42% → 91%).
|