4 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
Simon Layton	4e6a55751a	Force einsum to fp16	2019-10-14 11:12:41 -04:00
Simon Layton	899883644f	Fix test fails and warnings ... Attention output was in bnij ordering instead of ijbn which everything else will expect. This was an oversight on my part, and keeps the attention inputs/outputs identical to the original code. Also moved back from tensor slicing to index_select in rel_shift_bnij to make the tracer happy.	2019-10-03 12:05:15 -04:00
Simon Layton	9ffda216ec	Fix missed head transpose	2019-10-03 09:23:16 -04:00
Simon Layton	d51b589404	Re-order attention head outputs for better perf ... Significant performance boost over the original orderings on an already somewhat optimised branch this gave me > 2x end-to-end throughput on a squad xlnet fine-tuning task (batch 8, seq-length 612, fp16)	2019-10-02 12:18:21 -04:00