XXXXRT666
XXXXRT666
There could be in the future,and if you want it right now you could train it with g2p for those languages and larger datasets. You may refer to the approaches...
😅 😅 😅
api.py可以设置模型，有特定的endpoint，建议在同模型同参考下对比
不要把main分支和fast inference在同一环境下使用
语言选错了吧
AMD EPYC 7642 48C + 4090 Baseline 80it/s with BS 20 BS 20 225 it/s after compilation 190 it/s with CUDA Graph BS 1 900 it/s with CUDA Graph
update： **400it/s Batch Size 20** AMD EPYC 7642 48C + RTX 4090 with CUDA Graph and Flash Attention
logs中保存了，自取
git pull更新代码，还不行就opencc==xxx变成opencc