Stream Tiny Aya deltas through text route

app.py

@@ -467,6 +467,30 @@ def _tiny_aya_generate(system, user, max_tokens, temperature):
     return str(result or "")
 
 
+def _tiny_aya_stream(system, user, max_tokens, temperature):
+    from gradio_client import Client
+    client = Client(TINY_AYA_SPACE, token=HF_TOKEN or None)
+    try:
+        job = client.submit(
+            system or "",
+            user or "",
+            int(max_tokens or 400),
+            float(temperature if temperature is not None else 0.8),
+            api_name="/generate_stream",
+        )
+        prev = ""
+        for update in job:
+            text = update[0] if isinstance(update, (tuple, list)) else update
+            text = str(text or "")
+            if len(text) > len(prev):
+                yield text[len(prev):]
+            prev = text
+    except Exception:
+        text = _tiny_aya_generate(system, user, max_tokens, temperature)
+        if text:
+            yield text
+
+
 @fastapi_app.post("/voxcpm-tts")
 async def voxcpm_tts(request: Request):
     body = await request.json()
@@ -762,9 +786,10 @@ async def text_generate_stream(request: Request):
                 if model == "tiny-aya-global-zerogpu":
                     if not TINY_AYA_SPACE:
                         raise llm.LlmUnavailable("TINY_AYA_SPACE not set")
-                    text = _tiny_aya_generate(system, user, max_tokens, temperature)
-                    if text:
-                        loop.call_soon_threadsafe(q.put_nowait, ("delta", text))
+                    for chunk in _tiny_aya_stream(system, user, max_tokens, temperature):
+                        if stop.is_set():
+                            break
+                        loop.call_soon_threadsafe(q.put_nowait, ("delta", chunk))
                 else:
                     for chunk in llm.stream_chat(
                         system,