Updated config for min inference time

- This combination gives min inference time (~ 200 WPS)
   on local machine
This commit is contained in:
Abhishek Aggarwal 2021-02-16 14:21:46 +01:00
parent c5c5339489
commit 921c2eedf8

View File

@ -75,17 +75,25 @@ vocabs:
beam-size: 1
normalize: 1.0
word-penalty: 0
mini-batch: 32
maxi-batch: 100
maxi-batch-sort: src
max-input-sentence-tokens: 128
max-input-tokens: 1024
workspace: 128
max-length-factor: 2.0
skip-cost: true
cpu-threads: 1
quiet: true
quiet-translation: true
shortlist:
- /lex.${lang}.s2t
- 50
- 50
`;
/*
This config is not valid anymore in new APIs
mini-batch: 32
maxi-batch: 100
maxi-batch-sort: src
*/
// TODO: Use in model config when wormhole is enabled:
// gemm-precision: int8shift
// TODO: Use in model config when loading of binary models is supported and we use model.intgemm.alphas.bin: