Implement KV-ZCH Benchmark #3540
Closed
Conversation
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
Contributor
|
@Ali-Tehrani has exported this pull request. If you are a Meta employee, you can view the originating Diff in D86677315. |
meta-cla
bot
added
the
CLA Signed
Ali-Tehrani
force-pushed
the
export-D86677315
branch
from
89172a1 to
a67ac7a
Compare
Ali-Tehrani
pushed a commit
to Ali-Tehrani/torchrec
that referenced
this pull request
Nov 12, 2025
Summary:
Implements KV-ZCH with the benchmarking platform
Several things were added to make it work with KV-ZCH:
- Added eviction policies
- Added KeyValueParam to add parameters to TBE `fused_params` which is then fed into `SSDTableBatchedEmbeddingBags`. See `_populate_ssd_tbe_params` in batched_embedding_kernel and `add_params_from_parameter_sharding` in distributed/utils.py.
- Added CacheParams creation to set `prefetch_pipeline=True` due to warning below.
NOTE: The `prefetch_pipeline` attribute of `CacheParams` is set to True, due to the following complaint without it:
{F1983388476,width=300,height=200}
Update on November 11, 2025:
- The line `pipeline.progress(iter(bench_inputs)) ` is commented out on `benchmark_train_pipeline.py` due to conflict with `pipeline.reset()`. This gives an error on the forward pass when using `pipeline="prefetch"` with KV-ZCH.
Reviewed By: TroyGarden
Differential Revision: D86677315
Summary:
Implements KV-ZCH with the benchmarking platform
Several things were added to make it work with KV-ZCH:
- Added eviction policies
- Added KeyValueParam to add parameters to TBE `fused_params` which is then fed into `SSDTableBatchedEmbeddingBags`. See `_populate_ssd_tbe_params` in batched_embedding_kernel and `add_params_from_parameter_sharding` in distributed/utils.py.
- Added CacheParams creation to set `prefetch_pipeline=True` due to warning below.
NOTE: The `prefetch_pipeline` attribute of `CacheParams` is set to True, due to the following complaint without it:
{F1983388476,width=300,height=200}
Update on November 11, 2025:
- The line `pipeline.progress(iter(bench_inputs)) ` is commented out on `benchmark_train_pipeline.py` due to conflict with `pipeline.reset()`. This gives an error on the forward pass when using `pipeline="prefetch"` with KV-ZCH.
Reviewed By: TroyGarden
Differential Revision: D86677315
Ali-Tehrani
force-pushed
the
export-D86677315
branch
from
a67ac7a to
f21b537
Compare
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:
Implements KV-ZCH with the benchmarking platform
Several things were added to make it work with KV-ZCH:
fused_paramswhich is then fed intoSSDTableBatchedEmbeddingBags. See_populate_ssd_tbe_paramsin batched_embedding_kernel andadd_params_from_parameter_shardingin distributed/utils.py.prefetch_pipeline=Truedue to warning below.NOTE: The
prefetch_pipelineattribute ofCacheParamsis set to True, due to the following complaint without it:{F1983388476,width=300,height=200}
Update on November 11, 2025:
pipeline.progress(iter(bench_inputs))is commented out onbenchmark_train_pipeline.pydue to conflict withpipeline.reset(). This gives an error on the forward pass when usingpipeline="prefetch"with KV-ZCH.Reviewed By: TroyGarden
Differential Revision: D86677315