cyx-6 opened a new pull request, #14737: URL: https://github.com/apache/tvm/pull/14737
This PR expands the support of attention bias layout. In the past, we only support 3 types of layout, with `BNSS'`, `BSS'`, `BS'`, e.g. `BNSS'` for (batch_size, num_head, seq_len, seq_len_kv). However, cutlass accepts 8 types of layout, as long as the shape ends with `S` and is broadcastable to `BNSS'`. So in this PR, we fixed the ndim of `bias` to 4 and accepts all 8 types of layout as cutlass. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
