667 MB

Ctrl+K

Replace q4-mix encoder with q8 (dynamic_wi8_afp32): en 17.31->11.87 WER, zh 19.00->17.46 CER at same size (599 MB)

855f287 verified 5 days ago

.gitattributes

1.52 kB
initial commit 10 days ago
README.md

3.2 kB
model card: prompt-slot guidance — the FT repairs `auto` (base 50.58 -> 11.33 CER, best slot); all slots within ~0.7 CER 8 days ago
config.json

1.4 kB
v2: q4-mix LiteRT of the zh-TW FINE-TUNED model (CV zh-TW CER 13.90 vs ~38 for v1 warm-start; fp32 ref 12.03) 8 days ago
nemotron_decoder_fp16.tflite

30.7 MB
xet

v2: q4-mix LiteRT of the zh-TW FINE-TUNED model (CV zh-TW CER 13.90 vs ~38 for v1 warm-start; fp32 ref 12.03) 8 days ago
nemotron_encoder_q8.tflite

599 MB
xet

Replace q4-mix encoder with q8 (dynamic_wi8_afp32): en 17.31->11.87 WER, zh 19.00->17.46 CER at same size (599 MB) 5 days ago
nemotron_joint_fp16.tflite

18.1 MB
xet

v2: q4-mix LiteRT of the zh-TW FINE-TUNED model (CV zh-TW CER 13.90 vs ~38 for v1 warm-start; fp32 ref 12.03) 8 days ago
nemotron_prompt_fuse_fp32.tflite

17.8 MB
xet

v2: q4-mix LiteRT of the zh-TW FINE-TUNED model (CV zh-TW CER 13.90 vs ~38 for v1 warm-start; fp32 ref 12.03) 8 days ago
processor_config.json

2.52 kB
zh-TW variant v1.1: warm-started zh-TW slot (100%->15.78% CER), self-contained q4-mix build 10 days ago
tokenizer.json

752 kB
zh-TW variant v1.1: warm-started zh-TW slot (100%->15.78% CER), self-contained q4-mix build 10 days ago
tokenizer_config.json

881 Bytes
zh-TW variant v1.1: warm-started zh-TW slot (100%->15.78% CER), self-contained q4-mix build 10 days ago