Justin Wei

icon indicating a website link linkedin.com/in/JustinAWei

icon location Austin Area icon location Co-founder & CTO of Phind (YC S22)

Results 2 issues of Justin Wei

T5 cross_attention output cannot be accessed

1 comment

### Description As defined in the [fastertransformers T5 guide](https://github.com/NVIDIA/FasterTransformer/blob/main/docs/t5_guide.md) there is an output value for `cross_attentions`. I cannot find any way of returning `cross_attentions` on fastertransformers Triton backend for T5....

bug

ModelConfig.tokenizer_path and TrainConfig.save_best

### 📚 The doc issue ModelConfig.tokenizer_path has perhaps moved to something like TokenizerConfig? It is under `tokenizer:` block in the yml config files. Also, `save_best` under TrainConfig is not documented...

documentation