JinchuanTian/LongCap-35B-A3B
Audio-Text-to-Text • 35B • Updated • 43
LongCap
Hour-scale timestamped audio captioning and speech diarization: BF16 and FP8 checkpoints and the long-form benchmark.
Note BF16 weights (2x H100)
JinchuanTian/LongCap-35B-A3B-FP8
Audio-Text-to-Text • 35B • Updated • 44 • 1Note FP8 weights (1x H100); used for all reported scores
JinchuanTian/LongCapBench-TAC-Long
Updated • 35 • 1Note Human-annotated long-form benchmark: four full recordings (7.05 h, 6,520 events), source links, and saved model predictions