Spaces:

root404onion
/

sql-agent

Sleeping

App Files Files Community

root commited on Apr 5

Commit

38819be

1 Parent(s): 14de67e

hf

Browse files

Files changed (11) hide show

README.md +1 -0
__pycache__/inference.cpython-313.pyc +0 -0
core/__pycache__/__init__.cpython-313.pyc +0 -0
database/__pycache__/__init__.cpython-313.pyc +0 -0
database/__pycache__/sqlite_manager.cpython-313.pyc +0 -0
inference.py +10 -2
server/__pycache__/app.cpython-313.pyc +0 -0
server/__pycache__/environment.cpython-313.pyc +0 -0
server/app.py +6 -16
server/environment.py +9 -4
server/playground.html +477 -0

README.md CHANGED Viewed

@@ -94,6 +94,7 @@ This creates reproducible benchmark records for easy/medium/hard tasks.
 - `Dockerfile` is included and starts `uvicorn server.app:app`.
 - Space metadata is configured in this `README.md` frontmatter with `sdk: docker`.
 - Health endpoint: `/health`
 Local container run:
 ```bash

 - `Dockerfile` is included and starts `uvicorn server.app:app`.
 - Space metadata is configured in this `README.md` frontmatter with `sdk: docker`.
 - Health endpoint: `/health`
+- **Web playground:** open the Space URL (`/`). Click **Connect session** to open a WebSocket to `/ws`, then **Start episode (reset)** and **Run query (step)**. OpenEnv’s HTTP `POST /reset` and `POST /step` each use a fresh environment instance (stateless); the WebSocket session keeps one episode alive for the UI.
 Local container run:
 ```bash

__pycache__/inference.cpython-313.pyc CHANGED Viewed

Binary files a/__pycache__/inference.cpython-313.pyc and b/__pycache__/inference.cpython-313.pyc differ

core/__pycache__/__init__.cpython-313.pyc CHANGED Viewed

Binary files a/core/__pycache__/__init__.cpython-313.pyc and b/core/__pycache__/__init__.cpython-313.pyc differ

database/__pycache__/__init__.cpython-313.pyc CHANGED Viewed

Binary files a/database/__pycache__/__init__.cpython-313.pyc and b/database/__pycache__/__init__.cpython-313.pyc differ

database/__pycache__/sqlite_manager.cpython-313.pyc CHANGED Viewed

Binary files a/database/__pycache__/sqlite_manager.cpython-313.pyc and b/database/__pycache__/sqlite_manager.cpython-313.pyc differ

inference.py CHANGED Viewed

@@ -87,8 +87,16 @@ def main():
                 obs = self.target.reset()
                 return type('StepResult', (), {'observation': obs, 'reward': 0.0, 'done': False})()
             def step(self, action):
-                obs, reward, done = self.target.step(action)
-                return type('StepResult', (), {'observation': obs, 'reward': reward, 'done': done})()
             def close(self):
                 pass
         env = DirectClient(base_env)

                 obs = self.target.reset()
                 return type('StepResult', (), {'observation': obs, 'reward': 0.0, 'done': False})()
             def step(self, action):
+                obs = self.target.step(action)
+                return type(
+                    'StepResult',
+                    (),
+                    {
+                        'observation': obs,
+                        'reward': obs.reward if obs.reward is not None else 0.0,
+                        'done': obs.done,
+                    },
+                )()
             def close(self):
                 pass
         env = DirectClient(base_env)

server/__pycache__/app.cpython-313.pyc CHANGED Viewed

Binary files a/server/__pycache__/app.cpython-313.pyc and b/server/__pycache__/app.cpython-313.pyc differ

server/__pycache__/environment.cpython-313.pyc CHANGED Viewed

Binary files a/server/__pycache__/environment.cpython-313.pyc and b/server/__pycache__/environment.cpython-313.pyc differ

server/app.py CHANGED Viewed

@@ -1,3 +1,5 @@
 from openenv.core.env_server import create_fastapi_app
 from fastapi.responses import HTMLResponse
 import os
@@ -8,25 +10,13 @@ from server.environment import SqlEnvironment
 app = create_fastapi_app(SqlEnvironment, SqlAction, SqlObservation)
 @app.get("/")
 def root():
-    return HTMLResponse(
-        """
-        <html>
-          <head><title>SQL OpenEnv Space</title></head>
-          <body style="font-family: sans-serif; margin: 2rem;">
-            <h2>SQL Data Analyst OpenEnv Environment</h2>
-            <p>Environment is running.</p>
-            <ul>
-              <li>Health check: <a href="/health">/health</a></li>
-              <li>OpenAPI docs: <a href="/docs">/docs</a></li>
-            </ul>
-            <p>Use the OpenEnv client in <code>inference.py</code> to run baseline episodes.</p>
-          </body>
-        </html>
-        """
-    )
 @app.get("/health")

+from pathlib import Path
 from openenv.core.env_server import create_fastapi_app
 from fastapi.responses import HTMLResponse
 import os
 app = create_fastapi_app(SqlEnvironment, SqlAction, SqlObservation)
+_PLAYGROUND_HTML = Path(__file__).resolve().parent / "playground.html"
 @app.get("/")
 def root():
+    html = _PLAYGROUND_HTML.read_text(encoding="utf-8")
+    return HTMLResponse(html)
 @app.get("/health")

server/environment.py CHANGED Viewed

@@ -1,5 +1,4 @@
 import uuid
-from typing import Tuple
 from openenv.core.env_server import Environment
 from models import SqlAction, SqlObservation, SqlState, SqlReward
@@ -35,13 +34,17 @@ class SqlEnvironment(Environment):
             difficulty=first_task.difficulty_label,
         )
     def state(self) -> SqlState:
         return self._state
-    def step(self, action: SqlAction) -> Tuple[SqlObservation, float, bool]:
         self._state.step_count += 1
         query = action.query.strip()
         current_idx = self._state.current_task_index
         task = self.task_registry.get_task(current_idx)
         self._state.attempts_per_task[current_idx] += 1
@@ -135,6 +138,8 @@ class SqlEnvironment(Environment):
             execution_error=error_str,
             task_score=score,
             grader_feedback=grader_feedback,
         )
-        return obs, reward.total, done

 import uuid
 from openenv.core.env_server import Environment
 from models import SqlAction, SqlObservation, SqlState, SqlReward
             difficulty=first_task.difficulty_label,
         )
+    @property
     def state(self) -> SqlState:
         return self._state
+    def step(self, action: SqlAction) -> SqlObservation:
         self._state.step_count += 1
         query = action.query.strip()
         current_idx = self._state.current_task_index
+        n_tasks = self.task_registry.get_total_tasks()
+        while len(self._state.attempts_per_task) < n_tasks:
+            self._state.attempts_per_task.append(0)
         task = self.task_registry.get_task(current_idx)
         self._state.attempts_per_task[current_idx] += 1
             execution_error=error_str,
             task_score=score,
             grader_feedback=grader_feedback,
+            reward=reward.total,
+            done=done,
         )
+        return obs

server/playground.html ADDED Viewed

	@@ -0,0 +1,477 @@

+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1" />
+  <title>SQL Analyst OpenEnv · Playground</title>
+  <link rel="preconnect" href="https://fonts.googleapis.com" />
+  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
+  <link href="https://fonts.googleapis.com/css2?family=JetBrains+Mono:wght@400;600&family=Newsreader:ital,opsz,wght@0,6..72,400;0,6..72,700;1,6..72,400&display=swap" rel="stylesheet" />
+  <style>
+    :root {
+      --bg: #0f1419;
+      --surface: #1a222d;
+      --border: #2d3a4a;
+      --text: #e7ecf3;
+      --muted: #8b9cb3;
+      --accent: #3ddc97;
+      --warn: #f5a623;
+      --err: #ff6b6b;
+      --sql-bg: #121820;
+    }
+    * { box-sizing: border-box; }
+    body {
+      margin: 0;
+      min-height: 100vh;
+      font-family: "Newsreader", Georgia, serif;
+      background: var(--bg);
+      color: var(--text);
+      background-image:
+        radial-gradient(ellipse 120% 80% at 100% -20%, rgba(61, 220, 151, 0.08), transparent),
+        radial-gradient(ellipse 80% 50% at 0% 100%, rgba(45, 58, 74, 0.5), transparent);
+    }
+    .wrap {
+      max-width: 1180px;
+      margin: 0 auto;
+      padding: 1.75rem 1.25rem 3rem;
+    }
+    header {
+      display: flex;
+      flex-wrap: wrap;
+      align-items: baseline;
+      justify-content: space-between;
+      gap: 1rem;
+      margin-bottom: 1.5rem;
+      border-bottom: 1px solid var(--border);
+      padding-bottom: 1rem;
+    }
+    h1 {
+      font-size: 1.65rem;
+      font-weight: 700;
+      margin: 0;
+      letter-spacing: -0.02em;
+    }
+    .tag {
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.7rem;
+      text-transform: uppercase;
+      letter-spacing: 0.12em;
+      color: var(--accent);
+      border: 1px solid var(--border);
+      padding: 0.25rem 0.5rem;
+      border-radius: 4px;
+    }
+    .links a {
+      color: var(--muted);
+      font-size: 0.9rem;
+      margin-left: 1rem;
+    }
+    .links a:hover { color: var(--accent); }
+    .grid {
+      display: grid;
+      grid-template-columns: 1fr 1fr;
+      gap: 1rem;
+    }
+    @media (max-width: 900px) { .grid { grid-template-columns: 1fr; } }
+    .panel {
+      background: var(--surface);
+      border: 1px solid var(--border);
+      border-radius: 10px;
+      padding: 1rem 1.1rem;
+    }
+    .panel h2 {
+      font-size: 0.85rem;
+      font-weight: 700;
+      text-transform: uppercase;
+      letter-spacing: 0.06em;
+      color: var(--muted);
+      margin: 0 0 0.75rem;
+    }
+    .task-meta {
+      display: flex;
+      gap: 0.5rem;
+      flex-wrap: wrap;
+      margin-bottom: 0.75rem;
+    }
+    .pill {
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.72rem;
+      padding: 0.2rem 0.5rem;
+      border-radius: 4px;
+      background: var(--sql-bg);
+      border: 1px solid var(--border);
+    }
+    .pill.diff-easy { color: var(--accent); }
+    .pill.diff-medium { color: var(--warn); }
+    .pill.diff-hard { color: var(--err); }
+    pre.schema, .instruction {
+      font-size: 0.95rem;
+      line-height: 1.55;
+      white-space: pre-wrap;
+      word-break: break-word;
+      margin: 0;
+      color: var(--muted);
+    }
+    .instruction { color: var(--text); }
+    textarea#sql {
+      width: 100%;
+      min-height: 140px;
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.82rem;
+      line-height: 1.5;
+      background: var(--sql-bg);
+      color: var(--text);
+      border: 1px solid var(--border);
+      border-radius: 8px;
+      padding: 0.85rem;
+      resize: vertical;
+    }
+    textarea#sql:focus {
+      outline: none;
+      border-color: var(--accent);
+      box-shadow: 0 0 0 1px rgba(61, 220, 151, 0.25);
+    }
+    .actions {
+      display: flex;
+      flex-wrap: wrap;
+      gap: 0.6rem;
+      margin-top: 0.75rem;
+    }
+    button {
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.78rem;
+      padding: 0.55rem 1rem;
+      border-radius: 6px;
+      border: 1px solid var(--border);
+      cursor: pointer;
+      background: var(--surface);
+      color: var(--text);
+    }
+    button.primary {
+      background: rgba(61, 220, 151, 0.15);
+      border-color: var(--accent);
+      color: var(--accent);
+    }
+    button:hover { filter: brightness(1.12); }
+    button:disabled { opacity: 0.45; cursor: not-allowed; }
+    .metrics {
+      display: grid;
+      grid-template-columns: repeat(auto-fill, minmax(140px, 1fr));
+      gap: 0.5rem;
+      margin-top: 0.75rem;
+    }
+    .metric {
+      background: var(--sql-bg);
+      border: 1px solid var(--border);
+      border-radius: 6px;
+      padding: 0.5rem 0.65rem;
+    }
+    .metric label {
+      display: block;
+      font-size: 0.65rem;
+      text-transform: uppercase;
+      letter-spacing: 0.08em;
+      color: var(--muted);
+    }
+    .metric span {
+      font-family: "JetBrains Mono", monospace;
+      font-size: 1rem;
+    }
+    .error { color: var(--err); font-size: 0.9rem; margin-top: 0.5rem; }
+    .feedback { color: var(--warn); font-size: 0.9rem; margin-top: 0.35rem; }
+    table.result {
+      width: 100%;
+      border-collapse: collapse;
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.75rem;
+      margin-top: 0.5rem;
+    }
+    table.result th, table.result td {
+      border: 1px solid var(--border);
+      padding: 0.35rem 0.5rem;
+      text-align: left;
+    }
+    table.result th { background: var(--sql-bg); color: var(--muted); }
+    .log {
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.72rem;
+      line-height: 1.5;
+      max-height: 220px;
+      overflow-y: auto;
+      background: var(--sql-bg);
+      border: 1px solid var(--border);
+      border-radius: 8px;
+      padding: 0.65rem;
+      color: var(--muted);
+      white-space: pre-wrap;
+      word-break: break-all;
+    }
+    .full { grid-column: 1 / -1; }
+    .status-bar {
+      font-family: "JetBrains Mono", monospace;
+      font-size: 0.72rem;
+      color: var(--muted);
+      margin-top: 0.5rem;
+    }
+    .status-bar.ok { color: var(--accent); }
+    .status-bar.bad { color: var(--err); }
+  </style>
+</head>
+<body>
+  <div class="wrap">
+    <header>
+      <div>
+        <span class="tag">OpenEnv</span>
+        <h1>SQL data analyst playground</h1>
+      </div>
+      <div class="links">
+        <a href="/docs">API docs</a>
+        <a href="/health">Health</a>
+      </div>
+    </header>
+    <p class="status-bar" style="margin:-0.5rem 0 1.25rem;">
+      Interactive mode uses a persistent <strong>WebSocket</strong> session at <code>/ws</code>
+      (OpenEnv HTTP <code>/step</code> is stateless). Connect, then run reset → SQL steps.
+    </p>
+    <div class="grid">
+      <div class="panel">
+        <h2>Current task</h2>
+        <div class="task-meta">
+          <span id="taskId" class="pill">—</span>
+          <span id="difficulty" class="pill">—</span>
+        </div>
+        <p id="instruction" class="instruction">Click “Start episode” to load the first task.</p>
+      </div>
+      <div class="panel">
+        <h2>Schema</h2>
+        <pre id="schema" class="schema">—</pre>
+      </div>
+      <div class="panel full">
+        <h2>Your SQL</h2>
+        <textarea id="sql" placeholder="SELECT ..."></textarea>
+        <div class="actions">
+          <button type="button" class="primary" id="btnConnect">Connect session</button>
+          <button type="button" class="primary" id="btnReset" disabled>Start episode (reset)</button>
+          <button type="button" class="primary" id="btnStep" disabled>Run query (step)</button>
+          <button type="button" id="btnState" disabled>Refresh state</button>
+        </div>
+        <div id="reqStatus" class="status-bar">Disconnected — click “Connect session”.</div>
+      </div>
+      <div class="panel">
+        <h2>Last step</h2>
+        <div class="metrics">
+          <div class="metric"><label>Reward</label><span id="reward">—</span></div>
+          <div class="metric"><label>Task score</label><span id="score">—</span></div>
+          <div class="metric"><label>Episode done</label><span id="done">—</span></div>
+        </div>
+        <p id="graderFb" class="feedback" hidden></p>
+        <p id="execErr" class="error" hidden></p>
+      </div>
+      <div class="panel">
+        <h2>Result preview</h2>
+        <div id="resultTable"></div>
+      </div>
+      <div class="panel full">
+        <h2>Episode log</h2>
+        <div id="log" class="log"></div>
+      </div>
+    </div>
+  </div>
+  <script>
+    const $ = (id) => document.getElementById(id);
+    const logEl = $("log");
+    function log(msg) {
+      const t = new Date().toISOString().slice(11, 19);
+      logEl.textContent += "[" + t + "] " + msg + "\n";
+      logEl.scrollTop = logEl.scrollHeight;
+    }
+    function setDifficultyClass(el, d) {
+      el.className = "pill diff-" + (d || "").toLowerCase();
+    }
+    function renderObservation(obs) {
+      if (!obs) return;
+      $("taskId").textContent = obs.task_id || "—";
+      const diff = obs.difficulty || "";
+      const diffEl = $("difficulty");
+      diffEl.textContent = diff || "—";
+      setDifficultyClass(diffEl, diff);
+      $("instruction").textContent = obs.current_task_instruction || "";
+      $("schema").textContent = obs.schema_info || "—";
+    }
+    function renderTableFromJson(jsonStr) {
+      const host = $("resultTable");
+      host.innerHTML = "";
+      if (!jsonStr) {
+        host.textContent = "No rows yet.";
+        return;
+      }
+      let rows;
+      try { rows = JSON.parse(jsonStr); } catch (e) {
+        host.textContent = String(jsonStr).slice(0, 2000);
+        return;
+      }
+      if (!Array.isArray(rows) || rows.length === 0) {
+        host.textContent = "0 rows.";
+        return;
+      }
+      const cols = Object.keys(rows[0]);
+      const table = document.createElement("table");
+      table.className = "result";
+      const thead = document.createElement("thead");
+      const trh = document.createElement("tr");
+      cols.forEach((c) => {
+        const th = document.createElement("th");
+        th.textContent = c;
+        trh.appendChild(th);
+      });
+      thead.appendChild(trh);
+      table.appendChild(thead);
+      const tbody = document.createElement("tbody");
+      rows.slice(0, 50).forEach((row) => {
+        const tr = document.createElement("tr");
+        cols.forEach((c) => {
+          const td = document.createElement("td");
+          const v = row[c];
+          td.textContent = v === null || v === undefined ? "" : String(v);
+          tr.appendChild(td);
+        });
+        tbody.appendChild(tr);
+      });
+      table.appendChild(tbody);
+      host.appendChild(table);
+      if (rows.length > 50) {
+        const note = document.createElement("p");
+        note.className = "status-bar";
+        note.textContent = "Showing first 50 rows.";
+        host.appendChild(note);
+      }
+    }
+    function setStatus(el, ok, text) {
+      el.textContent = text;
+      el.className = "status-bar " + (ok ? "ok" : "bad");
+    }
+    let socket = null;
+    function setConnected(connected) {
+      $("btnReset").disabled = !connected;
+      $("btnStep").disabled = !connected;
+      $("btnState").disabled = !connected;
+      $("btnConnect").textContent = connected ? "Reconnect" : "Connect session";
+    }
+    function handleWsMessage(raw) {
+      let msg;
+      try {
+        msg = JSON.parse(raw);
+      } catch (e) {
+        log("parse error: " + e);
+        return;
+      }
+      const status = $("reqStatus");
+      if (msg.type === "error") {
+        setStatus(status, false, msg.data && msg.data.message ? msg.data.message : "Server error");
+        log("error: " + JSON.stringify(msg.data));
+        return;
+      }
+      if (msg.type === "observation") {
+        const pack = msg.data || {};
+        const obs = pack.observation;
+        const rew = pack.reward;
+        const done = pack.done;
+        setStatus(status, true, "OK");
+        log("observation reward=" + rew + " done=" + done);
+        $("reward").textContent = rew != null ? Number(rew).toFixed(3) : "—";
+        $("done").textContent = String(!!done);
+        $("score").textContent = obs && obs.task_score != null ? Number(obs.task_score).toFixed(3) : "—";
+        renderObservation(obs);
+        renderTableFromJson(obs && obs.execution_result);
+        const fb = $("graderFb");
+        if (obs && obs.grader_feedback) {
+          fb.textContent = obs.grader_feedback;
+          fb.hidden = false;
+        } else { fb.hidden = true; }
+        const er = $("execErr");
+        if (obs && obs.execution_error) {
+          er.textContent = obs.execution_error;
+          er.hidden = false;
+        } else { er.hidden = true; }
+        return;
+      }
+      if (msg.type === "state") {
+        setStatus(status, true, "State received (see log).");
+        log("state: " + JSON.stringify(msg.data).slice(0, 2000));
+        return;
+      }
+      log("unknown msg: " + raw.slice(0, 500));
+    }
+    function connectWs() {
+      if (socket && socket.readyState === WebSocket.OPEN) {
+        socket.close();
+      }
+      const proto = location.protocol === "https:" ? "wss:" : "ws:";
+      const url = proto + "//" + location.host + "/ws";
+      setStatus($("reqStatus"), true, "Connecting…");
+      socket = new WebSocket(url);
+      socket.onopen = function () {
+        setConnected(true);
+        setStatus($("reqStatus"), true, "WebSocket connected. Run reset to start.");
+        log("WebSocket open " + url);
+      };
+      socket.onclose = function () {
+        setConnected(false);
+        setStatus($("reqStatus"), false, "Disconnected.");
+        log("WebSocket closed");
+      };
+      socket.onerror = function () {
+        setStatus($("reqStatus"), false, "WebSocket error.");
+        log("WebSocket error");
+      };
+      socket.onmessage = function (ev) {
+        handleWsMessage(ev.data);
+      };
+    }
+    function wsSend(obj) {
+      if (!socket || socket.readyState !== WebSocket.OPEN) {
+        setStatus($("reqStatus"), false, "Not connected.");
+        return;
+      }
+      socket.send(JSON.stringify(obj));
+    }
+    $("btnConnect").addEventListener("click", connectWs);
+    $("btnReset").addEventListener("click", function () {
+      wsSend({ type: "reset", data: {} });
+      $("graderFb").hidden = true;
+      $("execErr").hidden = true;
+      renderTableFromJson(null);
+      log("sent reset");
+    });
+    $("btnStep").addEventListener("click", function () {
+      const q = $("sql").value.trim();
+      if (!q) {
+        setStatus($("reqStatus"), false, "Enter a SQL query.");
+        return;
+      }
+      wsSend({ type: "step", data: { query: q } });
+      log("sent step");
+    });
+    $("btnState").addEventListener("click", function () {
+      wsSend({ type: "state" });
+      log("sent state");
+    });
+    log("Open the Space, click Connect session, then Reset → run SQL.");
+  </script>
+</body>
+</html>