Fix runaway MLX Qwen audio chunks (#964)

* fix runaway MLX Qwen audio chunks * test: tighten runaway retry coverage --------- Co-authored-by: huanghua01 <[email protected]>
2026-09-18 06:10:43 -07:00 · 2026-07-26 23:31:23 -07:00
parent 624f6a2140
commit ed54347e81
7 changed files with 268 additions and 23 deletions
@@ -49,6 +49,7 @@ class ModelConfig:
    model_size: str = "default"
    size_mb: int = 0
    needs_trim: bool = False
+    retries_runaway: bool = False
    supports_instruct: bool = False
    languages: list[str] = field(default_factory=lambda: ["en"])
 ```
@@ -59,6 +60,7 @@ Registry helpers in `backends/__init__.py` replace what used to be per-engine `i
 - `get_tts_model_configs()` — only TTS variants
 - `get_model_config(model_name)` — lookup by name
 - `engine_needs_trim(engine)` — whether output should run through `trim_tts_output()`
+- `engine_retries_runaway(engine)` — whether unstable output should be retried as smaller chunks
 - `load_engine_model(engine, model_size)` — downloads + loads, handles engines with multiple sizes
 - `get_tts_backend_for_engine(engine)` — thread-safe backend factory with double-checked locking

@@ -152,7 +154,7 @@ The request path from frontend to audio file:

 6. **Inference** — the engine's `generate()` returns `(audio_array, sample_rate)`.

-7. **Post-process** — if `engine_needs_trim(engine)` is True, `trim_tts_output()` strips trailing silence. Effects chains (if any) are applied per generation version, not the clean version.
+7. **Validate and post-process** — engines with `retries_runaway=True` retry unstable output as smaller chunks. If `engine_needs_trim(engine)` is True, `trim_tts_output()` strips trailing silence. Effects chains (if any) are applied per generation version, not the clean version.

 8. **Persist** — audio is written to the generations directory, a row is inserted into the `generations` table, and the response includes the generation metadata.