Split fbpkg: source-only cea.chips.benchpress + self-contained cea.chips.benchpress.oss_prebuild (#686)#686
Open
excelle08 wants to merge 2 commits into
Conversation
meta-cla
Bot
added
the
CLA Signed
|
@excelle08 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D101889291. |
…arch#685) Summary: Pre-build FeedSim at fbpkg build time so the package runs without `benchpress install`. Both x86_64 and aarch64 use the same two-stage offline build: 1. `feedsim_deps_download` — runs on x86 with network, downloads source tarballs and pre-packed conda env from Manifold 2. `feedsim_oss_build[_aarch64]` — runs on arch-specific RE host, extracts conda env (GCC 15.2, cmake, ninja), builds all deps from source, compiles FeedSim with DLRM support Key changes: - `build.sh`: standalone build script — extracts pre-packed conda env, builds Boost/gflags/glog/jemalloc/libevent/folly/fbthrift, compiles FeedSim - `download_deps.sh`: downloads from Manifold (no internet fallback needed) - BUCK: three genrules (deps_download, x86 build, aarch64 build) with `select()` in fbpkg path_actions for architecture dispatch - CMake: pass-through CMAKE_PREFIX_PATH for staging dir in ExternalProject - `run.sh`/`run-feedsim-multi.sh`: LD_LIBRARY_PATH for conda runtime libs Performance (feedsim_autoscale, 3 iterations each): - x86 Bergamo: 395.20 QPS self-contained vs 380.92 QPS master (+3.7%) - ARM Grace: 324.57 QPS self-contained vs 322.48 QPS master (+0.6%) Reviewed By: gandhijayneel Differential Revision: D93702226
…ips.benchpress.oss_prebuild (facebookresearch#686) Summary: Split the DCPerf fbpkg into two packages: 1. `cea.chips.benchpress` — source-only package. Benchmarks must be installed on the target server via `benchpress install`. This restores the original behavior before pre-built FeedSim was added. 2. `cea.chips.benchpress.oss_prebuild` — self-contained package with pre-built benchmarks (currently FeedSim for x86_64 and aarch64). No `benchpress install` needed — just `fbpkg fetch` and `benchpress run`. Also switches LibTorch from PyTorch pre-built zips to conda-packed tarballs for both architectures, fixing ARM DLRM (no aarch64 libtorch zip from PyTorch) and updating motivation/context in the implementation plan doc. Reviewed By: gandhijayneel Differential Revision: D101889291
meta-codesync
Bot
changed the title
excelle08
force-pushed
the
export-D101889291-to-v2-beta
branch
from
2495e65 to
8394c3f
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
Jun 25, 2026
…ips.benchpress.oss_prebuild (#686) Summary: Pull Request resolved: #686 Split the DCPerf fbpkg into two packages: 1. `cea.chips.benchpress` — source-only package. Benchmarks must be installed on the target server via `benchpress install`. This restores the original behavior before pre-built FeedSim was added. 2. `cea.chips.benchpress.oss_prebuild` — self-contained package with pre-built benchmarks (currently FeedSim for x86_64 and aarch64). No `benchpress install` needed — just `fbpkg fetch` and `benchpress run`. Also switches LibTorch from PyTorch pre-built zips to conda-packed tarballs for both architectures, fixing ARM DLRM (no aarch64 libtorch zip from PyTorch) and updating motivation/context in the implementation plan doc. Reviewed By: gandhijayneel Differential Revision: D101889291 fbshipit-source-id: a022d75169fde9e43bf7dd89629378bf18939799
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
Split the DCPerf fbpkg into two packages:
cea.chips.benchpress— source-only package. Benchmarks must be installed on the target server viabenchpress install. This restores the original behavior before pre-built FeedSim was added.cea.chips.benchpress.oss_prebuild— self-contained package with pre-built benchmarks (currently FeedSim for x86_64 and aarch64). Nobenchpress installneeded — justfbpkg fetchandbenchpress run.Also switches LibTorch from PyTorch pre-built zips to conda-packed tarballs for both architectures, fixing ARM DLRM (no aarch64 libtorch zip from PyTorch) and updating motivation/context in the implementation plan doc.
Reviewed By: gandhijayneel
Differential Revision: D101889291