Hugging Face
Search models, datasets, users...
Ctrl K
Models
Datasets
Spaces
Buckets
Docs
Enterprise
Pricing
Website
Tasks
HuggingChat
Collections
Languages
Organizations
Community
Blog
Posts
Daily Papers
Hardware
Learn
Discord
Forum
GitHub
Solutions
Team & Enterprise
Hugging Face PRO
Enterprise Support
Inference Providers
Inference Endpoints
Storage Buckets
Log In
Sign Up
willwade
/
byt5-g2p-multilingual-tiny-kd
Like
0
PyTorch
ONNX
multilingual
t5
g2p
grapheme-to-phoneme
byt5
distillation
License:
apache-2.0
Model card
Files
Files and versions
xet
Community
Copy to bucket
main
byt5-g2p-multilingual-tiny-kd
163 MB
Ctrl+K
Ctrl+K
1 contributor
History:
6 commits
willwade
v3 fp32 ONNX refresh: Constant->initializer promoted, single-file (drops .onnx.data sidecar). Decode-gated.
9ffb80f
verified
3 days ago
onnx-int8
v3 int8 ONNX: properly quantized (dynamic QInt8, decode-gated). Fixes silent no-op quantization of Constant-node graphs.
3 days ago
onnx
v3 fp32 ONNX refresh: Constant->initializer promoted, single-file (drops .onnx.data sidecar). Decode-gated.
3 days ago
.gitattributes
Safe
1.52 kB
initial commit
3 days ago
README.md
Safe
1.34 kB
card: KD comparison artifact, scratch tiny is recommended
3 days ago
added_tokens.json
Safe
3.02 kB
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago
config.json
Safe
691 Bytes
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago
generation_config.json
Safe
112 Bytes
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago
pytorch_model.bin
71.8 MB
xet
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago
special_tokens_map.json
Safe
3.09 kB
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago
tokenizer_config.json
Safe
25.6 kB
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago
training_args.bin
5.78 kB
xet
v3 KD-tiny (distilled from g2p-small teacher): 66.2% micro / 56.7% macro. For comparison β the from-scratch tiny (69.3%) outperforms KD on G2P. The KD approach wins on P2G (the harder task) but loses on G2P.
3 days ago