Spaces:
Sleeping
Sleeping
File size: 7,238 Bytes
64ccc23 e864384 64ccc23 4cc4a08 64ccc23 4cc4a08 64ccc23 4cc4a08 64ccc23 4cc4a08 64ccc23 85465ba 64ccc23 | 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 | """
server/app.py
=============
FastAPI server for the Interactive SQL Query Repair environment.
Endpoints
---------
GET /health β liveness probe
POST /reset β start a new episode
POST /step β take one action
GET /state β current episode state
GET /tasks β list all available tasks
GET /grader β score the current (or last) episode
"""
from __future__ import annotations
import sys
import os
from fastapi import Body
from typing import Optional
# Allow imports from project root regardless of working directory
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from typing import Annotated, Union
from fastapi import FastAPI, HTTPException
from fastapi.responses import JSONResponse
from pydantic import BaseModel, Field
from app.data_loader import get_all_tasks, get_tasks_by_difficulty, task_count
from app.env import SQLRepairEnv
from app.models import (
AskClarificationAction,
Difficulty,
EnvState,
GraderResult,
HealthResponse,
Observation,
ResetRequest,
RunPartialAction,
StepResult,
SubmitQueryAction,
TaskListResponse,
TaskSummary,
)
# ---------------------------------------------------------------------------
# App setup
# ---------------------------------------------------------------------------
app = FastAPI(
title="Interactive SQL Query Repair Environment",
description=(
"An OpenEnv-compatible RL environment where an agent fixes broken SQL "
"queries through multi-turn interaction: running partial queries, asking "
"schema clarifications, and submitting final repairs."
),
version="1.0.0",
)
# One global environment instance per server process.
# For concurrent multi-agent evaluation, run multiple server processes.
_env = SQLRepairEnv()
# ---------------------------------------------------------------------------
# Action discriminator model
# ---------------------------------------------------------------------------
class StepRequest(BaseModel):
"""
Wrapper that discriminates between the three action types.
The 'type' field routes to the correct action model.
"""
type: str
query: str | None = None
question: str | None = None
def to_action(self) -> Union[SubmitQueryAction, RunPartialAction, AskClarificationAction]:
if self.type == "submit_query":
if not self.query:
raise ValueError("submit_query action requires 'query' field")
return SubmitQueryAction(query=self.query)
elif self.type == "run_partial":
if not self.query:
raise ValueError("run_partial action requires 'query' field")
return RunPartialAction(query=self.query)
elif self.type == "ask_clarification":
if not self.question:
raise ValueError("ask_clarification action requires 'question' field")
return AskClarificationAction(question=self.question)
else:
raise ValueError(
f"Unknown action type '{self.type}'. "
f"Must be one of: submit_query, run_partial, ask_clarification"
)
# ---------------------------------------------------------------------------
# Endpoints
# ---------------------------------------------------------------------------
@app.get("/health", response_model=HealthResponse, tags=["meta"])
def health() -> HealthResponse:
"""Liveness probe β always returns 200 if the server is running."""
return HealthResponse(status="ok")
@app.get("/tasks", response_model=TaskListResponse, tags=["meta"])
def list_tasks() -> TaskListResponse:
"""
List all available tasks across all difficulty levels.
Use the task 'id' field in POST /reset to start a specific episode.
"""
summaries = [
TaskSummary(
id=t.id,
difficulty=t.difficulty,
intent=t.intent,
error_type=t.error_type,
min_turns_expected=t.min_turns_expected,
)
for t in get_all_tasks()
]
return TaskListResponse(tasks=summaries)
@app.post("/reset", response_model=Observation, tags=["env"])
def reset(req: Optional[ResetRequest] = Body(default=None)) -> Observation:
try:
task_id = req.task_id if req and req.task_id else "easy_001"
difficulty = req.difficulty if req and req.difficulty else "easy"
obs = _env.reset(task_id=task_id, difficulty=difficulty)
return obs
except KeyError as e:
raise HTTPException(status_code=404, detail=str(e))
except ValueError as e:
raise HTTPException(status_code=422, detail=str(e))
except RuntimeError as e:
raise HTTPException(status_code=500, detail=str(e))
@app.post("/step", response_model=StepResult, tags=["env"])
def step(req: StepRequest) -> StepResult:
"""
Take one action in the current episode.
Body (one of three shapes)
--------------------------
Submit query (ends episode):
{ "type": "submit_query", "query": "SELECT ..." }
Run partial query (test without finalising):
{ "type": "run_partial", "query": "SELECT ..." }
Ask clarification:
{ "type": "ask_clarification", "question": "what is the join key?" }
Returns StepResult with observation, reward, done flag, and diagnostics.
Call GET /grader after done=true to get the final score.
"""
try:
action = req.to_action()
except ValueError as e:
raise HTTPException(status_code=422, detail=str(e))
try:
result = _env.step(action)
return result
except RuntimeError as e:
raise HTTPException(status_code=400, detail=str(e))
@app.get("/state", response_model=EnvState, tags=["env"])
def state() -> EnvState:
"""Return the current episode state (safe to call at any time)."""
return _env.state()
@app.get("/grader", response_model=GraderResult, tags=["eval"])
def grader() -> GraderResult:
"""
Compute and return the grader score for the current (or most recent) episode.
Score formula
-------------
final = 0.60 Γ correctness
+ 0.25 Γ turn_efficiency
+ 0.15 Γ clarification_quality
All sub-scores are in [0.0, 1.0].
Can be called mid-episode to see a partial score.
"""
return _env.grade()
# ---------------------------------------------------------------------------
# Error handlers
# ---------------------------------------------------------------------------
@app.exception_handler(Exception)
async def generic_exception_handler(request, exc):
return JSONResponse(
status_code=500,
content={"detail": f"Internal server error: {str(exc)}"},
)
import uvicorn
def main():
uvicorn.run("server.app:app", host="0.0.0.0", port=7860)
# ---------------------------------------------------------------------------
# Entry point (for direct python execution)
# ---------------------------------------------------------------------------
if __name__ == "__main__":
import uvicorn
uvicorn.run("server.app:app", host="0.0.0.0", port=7860, reload=False)
|