Research/alpa light#224

Draft

pgimenes wants to merge 98 commits intomainfrom

research/alpa-light

Commits on Jun 4, 2024

initial commit
Pedro Gimenes
committed

Commits on Jun 6, 2024

cvxpy bug for resharding cost term in intra operator pass
pgimenes
committed

Commits on Jun 11, 2024

Commits on Jun 12, 2024

Commits on Jun 13, 2024

Commits on Jun 14, 2024

Commits on Jun 19, 2024

pipeline for distributed inference + report parallelization pass
pgimenes
committed

Commits on Jun 25, 2024

layer norm and baddbmm ops
pgimenes
committed
[REFACTOR] Export strategies for each node using torch distributed format. Supported ops: mm, bmm, addmm, baddbmm
pgimenes
committed

Commits on Jun 28, 2024

Commits on Jul 3, 2024

Commits on Jul 10, 2024

Commits on Jul 11, 2024

Commits on Jul 15, 2024

support autosharding for OPT
pgimenes
committed

Commits on Jul 16, 2024

support activation modules, remove legacy stuff, some refactoring
pgimenes
committed

Commits on Jul 17, 2024

Commits on Jul 19, 2024

Commits on Jul 23, 2024

Commits on Jul 26, 2024

Commits on Jul 29, 2024

simplify mase launcher and get resharding nodes working with non fully replicated model
pgimenes
committed

Commits on Jul 30, 2024

Commits on Jul 31, 2024

remove logging which was slowing down runtime and remove redistribute nodes for non-tensor arguments
pgimenes
committed

Commits on Aug 5, 2024

DTensor: remove duplicated op call for out tensor meta propagation
pgimenes
committed

Commits on Aug 8, 2024

Commits on Aug 9, 2024

include DTensorCache to bypass DTensor construction + remove high overhead @torch._disable_dynamo decorator + clean up logs
pgimenes
committed

Commits on Aug 16, 2024

Commits on Aug 19, 2024

Commits on Aug 29, 2024

Commits on Aug 30, 2024

Commits on Sep 2, 2024

allgather/allreduce: replace cost db with regression model
pgimenes
committed

Commits on Sep 4, 2024

move all op benchmarking to use real vLLM classes
pgimenes
committed

Commits on Sep 5, 2024