Files
vllm-omni/tests
Sy03 92cecf0296 [Bugfix] Score complete long audio in Seed-TTS WER
Retain audio beyond the default Whisper 30-second window and enable long-form timestamp decoding. Extend the existing transcription regression with a 61-second case while preserving the short-audio path. Clarify the existing test loader and per-turn request copy for static checking.

Signed-off-by: Sy03 <1370724210@qq.com>
2026-09-08 17:40:59 +08:00
..