Feature: Consider SenseVoice for training data transcription

叠字会出现吞字

[Triton TRT-LLM] TRT engine max sequence length (3000 frames) too small for sentences >~40 words

ERROR

Short sentences glitches solved for me

Max sample sequence

NUM_WARMUP_UPDATES multiplied by number of GPU's during multi-GPU training

I'm learing from your project. I want sone help about custom dataset, vocabulary, and retrain from zero.

About training and finetune.

Stream-based resampling