Studio: hint at Model auto-switch in the OpenAI "No model loaded" 400 (#7006)

studio/backend/routes/inference.py

@@ -3344,6 +3344,21 @@ def _automatic_model_load_may_run() -> bool:
     return get_openai_auto_switch_enabled() or get_auto_unload_idle_seconds() > 0
 
 
+def _no_model_loaded_detail(base: str) -> str:
+    """Append a pointer to the opt-in auto-switch toggle to a "no model loaded"
+    error, but only when it's off. Auto-switch (default off) cold-loads a
+    requested downloaded GGUF, so an off toggle is the usual reason a request
+    naming a listed model still 400/503s; surface the fix. With it on the name
+    simply didn't resolve to a local GGUF, so the hint would mislead and is omitted."""
+    from utils.openai_auto_switch_settings import get_openai_auto_switch_enabled
+
+    if get_openai_auto_switch_enabled():
+        return base
+    return base + (
+        " Or enable Model auto-switch (Settings > API) to load a requested model automatically."
+    )
+
+
 async def _maybe_auto_switch_model(
     requested_model: Optional[str],
     fastapi_request: Request,
@@ -6451,7 +6466,7 @@ async def openai_chat_completions(
         if not backend.active_model_name:
             raise HTTPException(
                 status_code = 400,
-                detail = "No model loaded. Call POST /inference/load first.",
+                detail = _no_model_loaded_detail("No model loaded. Call POST /inference/load first."),
             )
         # Clean public id so the response never echoes a local path; the audio
         # branch below receives this sanitized label too.
@@ -9184,7 +9199,7 @@ async def openai_completions(request: Request, current_subject: str = Depends(ge
     if not llama_backend.is_loaded:
         raise HTTPException(
             status_code = 503,
-            detail = "No GGUF model loaded. Load a GGUF model first.",
+            detail = _no_model_loaded_detail("No GGUF model loaded. Load a GGUF model first."),
         )
     if not isinstance(body, dict):
         # Re-read to re-raise a malformed-body error (post-503, pre-feature behavior);
@@ -9400,7 +9415,7 @@ async def openai_embeddings(request: Request, current_subject: str = Depends(get
     if not llama_backend.is_loaded:
         raise HTTPException(
             status_code = 503,
-            detail = "No GGUF model loaded. Load a GGUF model first.",
+            detail = _no_model_loaded_detail("No GGUF model loaded. Load a GGUF model first."),
         )
     if not isinstance(body, dict):
         # Re-read to re-raise a malformed-body error (post-503, pre-feature behavior);
@@ -10140,7 +10155,7 @@ async def _responses_stream(
         # so the client sees a useful error instead of a dangling stream.
         raise HTTPException(
             status_code = 400,
-            detail = (
+            detail = _no_model_loaded_detail(
                 "Streaming /v1/responses requires a GGUF model loaded via "
                 "llama-server. Use non-streaming /v1/responses, "
                 "/v1/chat/completions, or load a GGUF model."
@@ -11379,7 +11394,7 @@ async def anthropic_count_tokens(
     if not llama_backend.is_loaded:
         raise HTTPException(
             status_code = 503,
-            detail = "No GGUF model loaded. Load a GGUF model first.",
+            detail = _no_model_loaded_detail("No GGUF model loaded. Load a GGUF model first."),
         )
 
     # Same Anthropic → OpenAI translation as anthropic_messages: system is
@@ -11452,7 +11467,7 @@ async def anthropic_messages(
     if not llama_backend.is_loaded and not _automatic_model_load_may_run():
         raise HTTPException(
             status_code = 503,
-            detail = "No GGUF model loaded. Load a GGUF model first.",
+            detail = _no_model_loaded_detail("No GGUF model loaded. Load a GGUF model first."),
         )
 
     # max_tokens is a required field on the Anthropic Messages API; real Anthropic
@@ -11503,7 +11518,7 @@ async def anthropic_messages(
     if not llama_backend.is_loaded:
         raise HTTPException(
             status_code = 503,
-            detail = "No GGUF model loaded. Load a GGUF model first.",
+            detail = _no_model_loaded_detail("No GGUF model loaded. Load a GGUF model first."),
         )
 
     # Advertised repo id after an auto-switch load, else a clean public id, never

studio/backend/tests/test_openai_auto_switch.py

@@ -3037,3 +3037,60 @@ def test_acquire_swap_gate_is_cancellation_safe():
         inference_route._auto_switch_process_lock.release()
 
     asyncio.run(asyncio.wait_for(main(), timeout = 5))
+
+
+def test_no_model_loaded_detail_appends_hint_only_when_off(monkeypatch):
+    # The "no model loaded" errors point at the opt-in auto-switch toggle so a
+    # request naming a listed-but-unloaded model is self-explanatory -- but only
+    # when it's off. With it on the name simply didn't resolve, so no hint.
+    base = "No GGUF model loaded. Load a GGUF model first."
+
+    monkeypatch.setattr(settings, "get_openai_auto_switch_enabled", lambda: False)
+    off = inference_route._no_model_loaded_detail(base)
+    assert off.startswith(base)
+    assert "Model auto-switch" in off and "Settings > API" in off
+
+    monkeypatch.setattr(settings, "get_openai_auto_switch_enabled", lambda: True)
+    assert inference_route._no_model_loaded_detail(base) == base
+
+
+def _run_responses_stream_no_model(monkeypatch, *, enabled, active_model_name):
+    # Drive _responses_stream's GGUF-not-loaded guard: llama backend unloaded,
+    # inference backend maybe holding a non-GGUF model. Returns the 400 detail.
+    from fastapi import HTTPException
+    from models.inference import ResponsesRequest, ChatMessage
+
+    monkeypatch.setattr(settings, "get_openai_auto_switch_enabled", lambda: enabled)
+    monkeypatch.setattr(
+        inference_route, "get_llama_cpp_backend", lambda: _FakeBackend(loaded_id = None)
+    )
+    monkeypatch.setattr(
+        inference_route,
+        "get_inference_backend",
+        lambda: type("_B", (), {"active_model_name": active_model_name})(),
+    )
+    payload = ResponsesRequest(model = "unsloth/Qwen3.5-4B-GGUF", stream = True)
+    messages = [ChatMessage(role = "user", content = "hi")]
+    with pytest.raises(HTTPException) as exc:
+        asyncio.run(inference_route._responses_stream(payload, messages, None))
+    assert exc.value.status_code == 400
+    return exc.value.detail
+
+
+def test_responses_stream_hint_matches_toggle_regardless_of_active_model(monkeypatch):
+    # Streaming /v1/responses shares the GGUF-only 400 with the other "no model
+    # loaded" sites, so the auto-switch hint attaches whenever the toggle is
+    # off -- including while a non-GGUF model is active, since auto-switch
+    # evicts it to load a resolved GGUF (_maybe_auto_switch_model's resolver
+    # branch has no active-model guard, unlike its reload-stash branch). Only
+    # the toggle being on suppresses it.
+    hinted = _run_responses_stream_no_model(monkeypatch, enabled = False, active_model_name = None)
+    assert "Model auto-switch" in hinted
+
+    on = _run_responses_stream_no_model(monkeypatch, enabled = True, active_model_name = None)
+    assert "Model auto-switch" not in on
+
+    non_gguf_loaded = _run_responses_stream_no_model(
+        monkeypatch, enabled = False, active_model_name = "unsloth/Llama-3.2-1B-Instruct"
+    )
+    assert "Model auto-switch" in non_gguf_loaded