trace: event tracer for the RPC backend, rpc-server and llama-server - #191

Draft

danielhanchen wants to merge 20 commits into

feature/pipeline-groupsfrom

feature/rpc-trace

Commits on Sep 5, 2026

trace: env gated event tracer for the RPC backend, rpc-server and llama-server
danielhanchen
committed
d7b1c24
trace: script for the traced layer split cells on the pair
danielhanchen
committed
d6693f8

Commits on Sep 6, 2026

trace: keep the GPU timing probe per backend registry and index the merge by time
danielhanchen
committed
0f221c8
cuda: map the event create, query and elapsed calls for HIP and MUSA
danielhanchen
committed
9b7c554
trace: flush the trace file periodically, and stop the peer rpc-server properly
danielhanchen
committed
6c4c6de
trace: attribute the idle time of a step to the host phase it fell in, drop the run artifacts
danielhanchen
committed
57d759d
Merge remote-tracking branch 'origin/feature/pipeline-groups' into feature/rpc-trace
danielhanchen
committed
abe16e4

Commits on Sep 7, 2026

trace: decompose one device's idle into scheduling and host stalls
danielhanchen
committed
988e09e

Commits on Sep 8, 2026

Commits on Sep 9, 2026