{"author":"chenyu","author_email":"chenyu@fastmail.com","author_time":1711082406,"commit_time":1711082406,"committer":"GitHub","committer_email":"noreply@github.com","hash":"82ce60e1723554ac5c07d85a10e3a2474e398898","message":"use JIT_BATCH_SIZE=4 for GPT2 3090 benchmark (#3870)\n\nsmaller first batch saves about 0.05 ms per token. 1.75ms / tok on local 3090","parents":["fe6ceff15f2e847e736a27c7cb774e51f42b1e7f"],"tree_hash":"2e481f705f56548a84fc19e29f3d510edda56345"}