feat: sglang backend support

1 files changed, 40 insertions, 0 deletions

diff --git a/app/tests/test_tts.py b/app/tests/test_tts.py
index 39b407d..7170262 100644
--- a/app/tests/test_tts.py
+++ b/app/tests/test_tts.py
@@ -30,6 +30,7 @@ from converter.clients import (
     BACKEND_AUDIOCPP,
     BACKEND_FASTER,
     BACKEND_QWEN,
+    BACKEND_SGLOMNI,
     LANGUAGE_CHOICES,
     LANGUAGE_ISO_CODES,
     MODEL_SIZE,
@@ -2496,6 +2497,19 @@ class BackendWiringTests(unittest.TestCase):
                 backend=BACKEND_AUDIOCPP, voice=voice,
                 instructions=instructions)
 
+    def _sglomni_converter(self, model="qwen3_tts_1_7b_base", clone=None,
+                           instructions=None):
+        from backends.sglomni.catalog import entry_by_key
+        api_url = "http://127.0.0.1:8100"
+        with patch("converter.converter.SgOmniTTSClient") as client:
+            client.return_value.entry = entry_by_key(model)
+            client.return_value.api_url = api_url
+            return AudiobookConverter(
+                voice_mode=VOICE_MODE_CLONE if clone else
+                (VOICE_MODE_DESIGN if instructions else VOICE_MODE_CUSTOM),
+                voice_clone_ref_audio=clone, backend=BACKEND_SGLOMNI,
+                model_id=model, instructions=instructions, api_url=api_url)
+
     def test_narrator_tag_uses_faster_voice_name(self):
         converter = self._faster_converter(voice="male_richard_poe")
         self.assertEqual(converter._narrator_tag(), "male_richard_poe")
@@ -2536,6 +2550,32 @@ class BackendWiringTests(unittest.TestCase):
             converter._print_banner()
         self.assertIn("higgs_audio_tts", buffer.getvalue())
 
+    def test_sglomni_banner_prints_model_and_resolves_model_id(self):
+        # Regression: the banner read self.model_id, which __init__ never
+        # stored — every sglomni run crashed there after a good connect.
+        converter = self._sglomni_converter(model="zonos2",
+                                            clone="voices/ref.wav")
+        self.assertEqual(converter.model_id, "zonos2")
+        buffer = io.StringIO()
+        with redirect_stdout(buffer):
+            converter._print_banner()  # must not raise
+        output = buffer.getvalue()
+        self.assertIn("ZONOS2", output)
+        self.assertIn("Zyphra/zonos2", output)
+        self.assertIn("voice cloning from a reference clip", output)
+
+    def test_sglomni_wiring_resolves_and_stores_the_model_key(self):
+        from backends.sglomni.catalog import entry_by_key
+        api_url = "http://127.0.0.1:8100"
+        with patch("converter.converter.SgOmniTTSClient") as client:
+            AudiobookConverter(voice_mode=VOICE_MODE_CLONE,
+                               voice_clone_ref_audio="voices/ref.wav",
+                               backend=BACKEND_SGLOMNI, model_id="zonos2",
+                               api_url=api_url)
+        self.assertEqual(
+            client.call_args.kwargs["model"], "zonos2")
+        self.assertEqual(entry_by_key("zonos2").repo, "Zyphra/zonos2")
+
     def test_non_faster_narrator_tag_unchanged(self):
         with tempfile.TemporaryDirectory() as tmp:
             ref = Path(tmp) / "ref.wav"