[CPU] Fixed BF16 Matmul inference precision #22994
Merged
Conversation
This file contains 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
dmitry-gorokhov
added
bug
maxnick
reviewed
Feb 21, 2024
src/plugins/intel_cpu/src/nodes/executors/fullyconnected_implementations.cpp
Outdated
Show resolved
Hide resolved
dmitry-gorokhov
force-pushed
the
fix/bf16_matmul_on_fp32_ir
branch
from
eb634c4 to
bbb231f
Compare
maxnick
approved these changes
Feb 22, 2024
github-merge-queue bot
pushed a commit
that referenced
this pull request
Feb 22, 2024
CPU plugin uses EnforceInferencePrecision routine for BF16 precision mark-up. Its logic assumes only activations precision is changed before Matmul op, while weights precision keeps w/o any changes. Since dnnlFCTypeMapping misses BF16 activation, FP32 weights optimized configuration for bf16, execution always happens in FP32 precision even user manually set infer_precision=bf16. This bug is not visible on FP16 IRs (since BF16+FP16 config is present), so only FP32 IRs affected. SInce save_model and ovc apply FP16 compression be default, the issue mostly applicable for pipelines which use a model directly after convert_model call. Cherry-picks: #22994
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Details:
CPU plugin uses EnforceInferencePrecision routine for BF16 precision mark-up. Its logic assumes only activations precision is changed before Matmul op, while weights precision keeps w/o any changes. Since dnnlFCTypeMapping misses BF16 activation, FP32 weights optimized configuration for bf16, execution always happens in FP32 precision even user manually set infer_precision=bf16.
This bug is not visible on FP16 IRs (since BF16+FP16 config is present), so only FP32 IRs affected. SInce
save_modeland ovc apply FP16 compression be default, the issue mostly applicable for pipelines which use a model directly afterconvert_modelcall.