Spaces:

tandevllc
/

axis

Sleeping

App Files Files Community

smridhgupta commited on Nov 5, 2025

Commit

4877523

verified ·

1 Parent(s): 929dd6c

Update server.py

Browse files

Files changed (1) hide show

server.py +10 -135

server.py CHANGED Viewed

@@ -8,7 +8,7 @@ Gemini-CLI style planner + executor:
  • robust JSON extraction (balanced braces) to avoid parse failures
 """
-from flask import Flask, request, jsonify, Response
 from huggingface_hub import snapshot_download
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 import subprocess, os, json, traceback, io, contextlib
@@ -33,7 +33,7 @@ MODEL_ID = os.environ.get("MODEL_ID", "TinyLlama/TinyLlama-1.1B-Chat-v1.0")
 # ──────────────────────────────────────────────
 #  3) Flask app & actions executor
 # ──────────────────────────────────────────────
-app = Flask(__name__)
 SERVER_OS = platform.system().lower()
 ALLOW_AUTO_INSTALL = os.environ.get("ALLOW_AUTO_INSTALL", "0") == "1"
@@ -1689,144 +1689,19 @@ def assist_rewrite():
     out = llm_generate_text_exact(model, sys, user, max_new_tokens={"short":400,"medium":1200,"long":2400,"xl":4800}[length])
     return jsonify({"new_content": _sanitize_generated_content(None, "text", out)})
-HOME_HTML = Template(r"""<!doctype html>
-<html lang="en">
-<head>
-<meta charset="utf-8"/>
-<meta name="viewport" content="width=device-width,initial-scale=1"/>
-<title>Axis · Llama3 Agent API</title>
-<style>
-  :root { --bg:#0b0f17; --fg:#e6eef8; --muted:#a9b7c6; --card:#121825; --accent:#7aa2f7; --ok:#38d39f; --bad:#ff6b6b; }
-  * { box-sizing:border-box; }
-  body { margin:0; font:15px/1.55 system-ui, -apple-system, Segoe UI, Roboto, Ubuntu, Cantarell, 'Helvetica Neue', Arial; background:radial-gradient(1200px 600px at 10% -10%, #122034, transparent), var(--bg); color:var(--fg); }
-  header { padding:36px 20px 10px; max-width:980px; margin:0 auto; }
-  .badge { display:inline-block; padding:4px 10px; border:1px solid #2c3650; border-radius:999px; font-size:12px; color:var(--muted); }
-  h1 { margin:12px 0 8px; font-size:28px; letter-spacing:.2px; }
-  .sub { color:var(--muted); margin:0 0 18px; }
-  .grid { display:grid; grid-template-columns: 1.1fr 1fr; gap:18px; max-width:980px; padding:0 20px 36px; margin:0 auto; }
-  .card { background:linear-gradient(180deg, rgba(255,255,255,0.02), rgba(255,255,255,0.00)); border:1px solid #1a2336; border-radius:14px; padding:16px; }
-  .card h3 { margin:0 0 8px; font-size:16px; }
-  pre, code, textarea { font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, "Liberation Mono", monospace; }
-  pre { background:#0e1422; color:#cfe3ff; padding:12px; border-radius:10px; overflow:auto; border:1px solid #1a2336; }
-  .row { display:flex; gap:10px; align-items:center; flex-wrap:wrap; }
-  textarea { width:100%; min-height:120px; padding:10px; border-radius:10px; border:1px solid #202a40; background:#0e1422; color:#e6eef8; }
-  input[type="password"] { width:100%; padding:10px; border-radius:10px; border:1px solid #202a40; background:#0e1422; color:#e6eef8; }
-  button { background:var(--accent); color:#0b0f17; border:none; padding:10px 14px; border-radius:10px; cursor:pointer; font-weight:600; }
-  button:disabled { filter:grayscale(.5); opacity:.7; cursor:not-allowed; }
-  .muted { color:var(--muted); }
-  footer { max-width:980px; margin:0 auto 32px; padding:0 20px; color:var(--muted); }
-  a { color:#9ec1ff; text-decoration:none; }
-  .ok { color:var(--ok); }
-</style>
-</head>
-<body>
-  <header>
-    <span class="badge">Axis · Llama3 Agent API</span>
-    <h1>Welcome 👋</h1>
-    <p class="sub">This Space exposes a tiny agent server powered by <b>$MODEL</b>.
-    Use it with our CLI client, or hit the HTTP endpoints directly.</p>
-    <div class="row muted">
-      <div>Live status: <span class="ok">Running</span></div>
-      <div>•</div>
-      <div>Endpoints: <code>/infer</code> <code>/execute</code> <code>/gen</code> <code>/status</code></div>
-    </div>
-  </header>
-  <section class="grid">
-    <div class="card">
-      <h3>Quick start (CLI)</h3>
-      <p class="muted">Point the client at this Space and set an API key (if enabled).</p>
-      <pre>export LLAMA_SERVER="https://tandevllc-axis.hf.space"
-export LLAMA_API_KEY="&lt;your-key&gt;"   # optional
-python3 client.py</pre>
-      <h3>HTTP examples (cURL)</h3>
-      <pre>curl -X POST "$LLAMA_SERVER/infer" \
-  -H "content-type: application/json" \
-  -H "authorization: Bearer &lt;your-key&gt;" \
-  -d '{"prompt":"List three things this API can do"}'</pre>
-      <pre>curl -X POST "$LLAMA_SERVER/execute" \
-  -H "content-type: application/json" \
-  -H "authorization: Bearer &lt;your-key&gt;" \
-  -d '{"plan":{"steps":[{"type":"respond","text":"Hello from Axis"}]}}'</pre>
-      <pre>curl -X POST "$LLAMA_SERVER/gen" \
-  -H "content-type: application/json" \
-  -H "authorization: Bearer &lt;your-key&gt;" \
-  -d '{"format":"text","instruction":"one-line haiku about security"}'</pre>
-    </div>
-    <div class="card">
-      <h3>Try it here</h3>
-      <p class="muted">Send a request to <code>/infer</code> from the browser.</p>
-      <label class="muted">API key (optional)</label>
-      <input id="key" type="password" placeholder="Bearer token if required"/>
-      <label class="muted" style="margin-top:8px; display:block;">Prompt</label>
-      <textarea id="prompt">Say hello in one short line.</textarea>
-      <div class="row" style="margin-top:10px;">
-        <button id="run">Run</button>
-        <span id="status" class="muted"></span>
-      </div>
-      <pre id="out" style="margin-top:10px; min-height:140px;">(response appears here)</pre>
-    </div>
-  </section>
-  <section class="grid" style="grid-template-columns: 1fr;">
-    <div class="card">
-      <h3>Endpoint contract</h3>
-      <ul class="muted">
-        <li><code>POST /infer</code> → <code>{"plan":{"steps":[...]}}</code> (first call may be slow while the Space warms up)</li>
-        <li><code>POST /execute</code> with <code>{"plan": ...}</code> → <code>{"results":[...]}</code></li>
-        <li><code>POST /gen</code> with <code>{"format","instruction","length"}</code> → <code>{"content":"..."}</code></li>
-        <li><code>GET /status</code> → JSON status</li>
-        <li><span class="muted">Send <code>Authorization: Bearer &lt;token&gt;</code> if your server enforces API keys.</span></li>
-      </ul>
-    </div>
-  </section>
-  <footer>
-    <p>Built by TanDev · Axis server · Model: <b>$MODEL</b>. For programmatic checks, hit <a href="/status">/status</a> or request <code>Accept: application/json</code> on <code>/</code>.</p>
-  </footer>
-<script>
-  const $ = (id)=>document.getElementById(id);
-  $("run").addEventListener("click", async () => {
-    $("run").disabled = true; $("status").textContent = "Calling /infer…";
-    $("out").textContent = "";
-    try {
-      const headers = { "content-type": "application/json" };
-      const key = $("key").value.trim();
-      if (key) headers["authorization"] = key.startsWith("Bearer ") ? key : ("Bearer " + key);
-      const r = await fetch("/infer", {
-        method:"POST", headers, body: JSON.stringify({ prompt: $("prompt").value })
-      });
-      const txt = await r.text();
-      $("out").textContent = txt;
-      $("status").textContent = r.ok ? "OK" : ("HTTP " + r.status);
-    } catch (e) {
-      $("status").textContent = "Error";
-      $("out").textContent = String(e);
-    } finally {
-      $("run").disabled = false;
-    }
-  });
-</script>
-</body>
-</html>""")
 @app.get("/")
-def home():
-    # Keep JSON for probes/clients
-    accept = (request.headers.get("accept") or "").lower()
-    if "application/json" in accept or request.args.get("format") == "json":
-        return jsonify(ok=True, model=MODEL_NAME, endpoints=["/infer", "/execute", "/gen", "/status"])
-    return Response(HOME_HTML.substitute(MODEL=MODEL_NAME), mimetype="text/html")
 @app.get("/status")
 def status():
     return jsonify(ok=True, model=MODEL_NAME)
 # ──────────────────────────────────────────────
 #  4) Main
 # ──────────────────────────────────────────────

  • robust JSON extraction (balanced braces) to avoid parse failures
 """
+from flask import Flask, request, jsonify, Response, send_from_directory
 from huggingface_hub import snapshot_download
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 import subprocess, os, json, traceback, io, contextlib
 # ──────────────────────────────────────────────
 #  3) Flask app & actions executor
 # ──────────────────────────────────────────────
+app = Flask(__name__, static_folder="public", static_url_path="")
 SERVER_OS = platform.system().lower()
 ALLOW_AUTO_INSTALL = os.environ.get("ALLOW_AUTO_INSTALL", "0") == "1"
     out = llm_generate_text_exact(model, sys, user, max_new_tokens={"short":400,"medium":1200,"long":2400,"xl":4800}[length])
     return jsonify({"new_content": _sanitize_generated_content(None, "text", out)})
 @app.get("/")
+def root():
+    return send_from_directory(app.static_folder, "index.html")
+# (optional convenience)
+@app.get("/index.html")
+def root_html():
+    return send_from_directory(app.static_folder, "index.html")
+# keep status JSON as-is
 @app.get("/status")
 def status():
     return jsonify(ok=True, model=MODEL_NAME)
 # ──────────────────────────────────────────────
 #  4) Main
 # ──────────────────────────────────────────────