Seunghun Ji
Results
2
issues of
Seunghun Ji
Hi, first of all, thank you very much for this wonderful work. Recently I encountered an issue that, in seq2seq's cross-attention scenario where `flash_attn_varlen_kvpacked_func()` is used for training and `flash_attn_with_kvcache()`...
**Describe the bug** Hi, I recently started to use the Spanish text normalizer, and I found a bug. I don't expect the normalizer to convert the conjunction 'o' into another...
bug