Migration From Client #2
Conversation
fpetrini15
requested review from
mc-nv,
debermudez,
matthewkotila and
nvda-mesharma
fpetrini15
force-pushed
the
fpetrini-migrate-pa
branch
from
cc8b7cd to
6506d36
Compare
fpetrini15
removed request for
debermudez,
matthewkotila,
mc-nv and
nvda-mesharma
| PUBLIC TRITON_ENABLE_PERF_ANALYZER_OPENAI=1 | ||
| ) | ||
| if(NOT(${TRITON_ENABLE_CC_HTTP} AND ${TRITON_ENABLE_CC_GRPC})) | ||
| message(FATAL_ERROR "perf_analyzer requires both http and grpc client libraries.") |
There was a problem hiding this comment.
Is this a relic from the previous build? PA can set the protocol to be GRPC or HTTP or even just CAPI. I wonder if we can make these optional in the future, if they are not now.
There was a problem hiding this comment.
When I first started working on this my understanding from a huddle I had was that PA requires both the GRPC and HTTP client. Is this no longer the case? If so, I can cut a ticket for further build optimization.
There was a problem hiding this comment.
CC: Deferring to @matthewkotila. It's possible that PA requires both. If so, we may want to look at letting PA have HTTP, GRPC, or C-API (or any combination).
| grpcclient_static | ||
| httpclient_static | ||
| triton-core-serverapi # from repo-core | ||
| PRIVATE |
There was a problem hiding this comment.
Similar general question to above: will these files be duplicated across repos? In the future, will we need to make sure to mirror the changes in a CMakeLists in the clients/server repo here?
There was a problem hiding this comment.
See answer above.
Goal: Seamlessly move PA from client to its own repository.
Client changes: triton-inference-server/client#722