List your company
ResearchFormerly WizWand

Mandarin Dialect Identification on KeSpeech (test)

78.77AccuracyLoRA
55.442861.498967.55573.6111Nov 30, 2025
Updated 8mo ago

Evaluation results

MethodLinks
LoRARank (r)=256, Trainable Ratio (%)=9.85, GPU Mem. (GiB)=54.95, Time/epoch (min)=16.092025.1178.77
Vanilla Fine-TuningTrainable Ratio (%)=96.48, GPU Mem. (GiB)=76.14, Time/epoch (min)=16.582025.1178.67
LoRARank (r)=128, Trainable Ratio (%)=5.28, GPU Mem. (GiB)=53.76, Time/epoch (min)=15.992025.1178.37
LoRARank (r)=64, Trainable Ratio (%)=2.82, GPU Mem. (GiB)=53.14, Time/epoch (min)=15.922025.1178.06
LSTReduction Factor (RF)=8, Trainable Ratio (%)=2.06, GPU Mem. (GiB)=20.39, Time/epoch (min)=8.582025.1178
AdaLoRAInitial Rank (init_r)=256, Trainable Ratio (%)=9.86, GPU Mem. (GiB)=56.87, Time/epoch (min)=18.492025.1177.8
AdaLoRAInitial Rank (init_r)=128, Trainable Ratio (%)=5.29, GPU Mem. (GiB)=53.44, Time/epoch (min)=18.352025.1177.71
UniPTReduction Factor (RF)=4, Trainable Ratio (%)=2.85, GPU Mem. (GiB)=36.76, Time/epoch (min)=16.252025.1177.58
UniPTReduction Factor (RF)=2, Trainable Ratio (%)=5.31, GPU Mem. (GiB)=43.32, Time/epoch (min)=17.402025.1177.47
LSTReduction Factor (RF)=2, Trainable Ratio (%)=7.09, GPU Mem. (GiB)=28.37, Time/epoch (min)=9.922025.1177.37
LSTReduction Factor (RF)=4, Trainable Ratio (%)=3.80, GPU Mem. (GiB)=22.88, Time/epoch (min)=9.172025.1177.18
AdaLoRAInitial Rank (init_r)=64, Trainable Ratio (%)=2.82, GPU Mem. (GiB)=52.87, Time/epoch (min)=18.312025.1176.86
UniPTReduction Factor (RF)=8, Trainable Ratio (%)=1.57, GPU Mem. (GiB)=34.26, Time/epoch (min)=15.512025.1176.43
AdapterHidden Dimension (dim)=64, Trainable Ratio (%)=2.88, GPU Mem. (GiB)=64.21, Time/epoch (min)=16.212025.1176.26
SHERLReduction Factor (RF)=2, Trainable Ratio (%)=7.92, GPU Mem. (GiB)=45.08, Time/epoch (min)=10.782025.1175.98
SHERLReduction Factor (RF)=8, Trainable Ratio (%)=1.54, GPU Mem. (GiB)=21.54, Time/epoch (min)=8.022025.1175.88
AdapterHidden Dimension (dim)=128, Trainable Ratio (%)=5.34, GPU Mem. (GiB)=65.34, Time/epoch (min)=16.332025.1175.83
AdapterHidden Dimension (dim)=256, Trainable Ratio (%)=9.91, GPU Mem. (GiB)=67.34, Time/epoch (min)=16.502025.1175.54
SHERLReduction Factor (RF)=4, Trainable Ratio (%)=3.30, GPU Mem. (GiB)=31.64, Time/epoch (min)=9.032025.1175.43
BitFitTrainable Ratio (%)=0.33, GPU Mem. (GiB)=47.33, Time/epoch (min)=15.052025.1168.83
ResNet-342025.1161.13
ECAPA-TDNN2025.1160.77
Head TuningTrainable Ratio (%)=0.22, GPU Mem. (GiB)=7.06, Time/epoch (min)=5.682025.1159.33
Kaldi-xvector2025.1156.34