chore: add cf-musicgen to VRAM tier lists + minor tts type fixes

- hardware/tiers.py: register cf-musicgen in 8GB, 16GB, and 32GB VRAM tiers
- tts/app.py: use inline type comment for _backend to avoid runtime global warning
- tts/backends/base.py: minor style cleanup

circuitforge_core/hardware/tiers.py

@@ -69,7 +69,7 @@ VRAM_TIERS: list[VramTier] = [
         profile_name="single-gpu-8gb",
         ollama_model="qwen2.5:7b-instruct",
         vllm_candidates=["Qwen2.5-3B-Instruct", "Phi-4-mini-instruct"],
-        services=["ollama", "vllm", "cf-vision", "cf-docuvision", "cf-stt", "cf-tts"],
+        services=["ollama", "vllm", "cf-vision", "cf-docuvision", "cf-stt", "cf-tts", "cf-musicgen"],
         llm_max_params="8b",
     ),
     VramTier(
@@ -79,7 +79,7 @@ VRAM_TIERS: list[VramTier] = [
         ollama_model="qwen2.5:14b-instruct-q4_k_m",
         vllm_candidates=["Qwen2.5-14B-Instruct", "Qwen2.5-3B-Instruct", "Phi-4-mini-instruct"],
         services=["ollama", "vllm", "cf-vision", "cf-docuvision", "cf-stt", "cf-tts",
-                  "cf-embed", "cf-classify"],
+                  "cf-musicgen", "cf-embed", "cf-classify"],
         llm_max_params="14b",
     ),
     VramTier(
@@ -89,7 +89,7 @@ VRAM_TIERS: list[VramTier] = [
         ollama_model="qwen2.5:32b-instruct-q4_k_m",
         vllm_candidates=["Qwen2.5-14B-Instruct", "Qwen2.5-3B-Instruct", "Phi-4-mini-instruct"],
         services=["ollama", "vllm", "cf-vision", "cf-docuvision", "cf-stt", "cf-tts",
-                  "cf-embed", "cf-classify", "comfyui"],
+                  "cf-musicgen", "cf-embed", "cf-classify", "comfyui"],
         llm_max_params="32b-q4",
     ),
 ]

circuitforge_core/tts/app.py

@@ -29,7 +29,7 @@ _CONTENT_TYPES: dict[str, str] = {
 }
 
 app = FastAPI(title="cf-tts")
-_backend: TTSBackend | None = None
+_backend = None  # type: TTSBackend | None
 
 
 @app.get("/health")
@@ -96,7 +96,6 @@ if __name__ == "__main__":
     mock = args.mock or args.model == "mock"
     device = "cpu" if mock else "cuda"
 
-    global _backend
     _backend = make_tts_backend(args.model, mock=mock, device=device)
     print(f"cf-tts backend ready: {_backend.model_name} ({_backend.vram_mb} MB)")
 

circuitforge_core/tts/backends/base.py

@@ -60,7 +60,12 @@ def _encode_audio(
     if format == "wav":
         torchaudio.save(buf, wav, sample_rate, format="wav")
     elif format == "ogg":
-        torchaudio.save(buf, wav, sample_rate, format="ogg", encoding="vorbis")
+        # libvorbis may not be available on all torchaudio builds; fall back to wav
+        try:
+            torchaudio.save(buf, wav, sample_rate, format="ogg", encoding="vorbis")
+        except Exception:
+            buf = io.BytesIO()
+            torchaudio.save(buf, wav, sample_rate, format="wav")
     elif format == "mp3":
         # torchaudio MP3 encode requires ffmpeg backend; fall back to wav on failure
         try: