[LPT] FP8 KV-Cache Static Quantization #31557
Open
+863
−124
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
github-actions
bot
added
category: IE Tests
v-Golubev
commented
Aug 1, 2025
Comment on lines
+556
to
+575
| match_multi_query_bcst(const std::shared_ptr<ov::Node>& kv) { | ||
| using namespace ov::pass; | ||
| using namespace ov::pass::pattern; | ||
| using namespace ov::gen_pattern; | ||
|
|
||
| auto reshape_kv = wrap_type<ov::op::v1::Reshape>({kv, any_input()}); | ||
| auto unsqueeze_kv = wrap_type<ov::op::v0::Unsqueeze>({kv, any_input()}); | ||
|
|
||
| auto constant_bcst = wrap_type<ov::op::v0::Constant>(value_matches("1.0")); | ||
|
|
||
| auto computed_bcst = makePattern<ov::op::v1::Broadcast>( | ||
| {wrap_type<ov::op::v0::Constant>(value_matches("1.0")), any_input(), any_input()}, | ||
| {{"mode", "numpy"}}); | ||
|
|
||
| auto multiply_kv = wrap_type<ov::op::v1::Multiply>({reshape_kv | unsqueeze_kv, constant_bcst | computed_bcst}); | ||
| auto computed_bcst3 = makePattern<ov::op::v3::Broadcast>({unsqueeze_kv, any_input()}, {{"mode", "bidirectional"}}); | ||
|
|
||
| auto result = wrap_type<ov::op::v1::Reshape>({multiply_kv | computed_bcst3, any_input()}); | ||
| return std::make_tuple(result, reshape_kv, unsqueeze_kv, computed_bcst, multiply_kv, computed_bcst3); | ||
| } |
There was a problem hiding this comment.
TODO: rewrite after #31386 is merged into master
aobolensk
reviewed
Aug 1, 2025
| auto fc_cache = fake_convert->clone_with_new_inputs(form_new_fc_inputs(concat->input_value(0))); | ||
| fc_cache->set_friendly_name(fake_convert->get_friendly_name() + "_1"); | ||
| auto fc_kv = fake_convert->clone_with_new_inputs(form_new_fc_inputs(concat->input_value(1))); | ||
| fc_cache->set_friendly_name(fake_convert->get_friendly_name() + "_2"); |
There was a problem hiding this comment.
Did you mean this?
Suggested change
| fc_cache->set_friendly_name(fake_convert->get_friendly_name() + "_2"); | |
| fc_kv->set_friendly_name(fake_convert->get_friendly_name() + "_2"); |
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:
This PR adds KV-Cache static quantization support on LPT side. The main changes:
MoveFakeConvertUpThroughKVCacheConcattransformation: a prerequisite for KV cache quantization optimization. It identifies patterns whereFakeConvertoperations are applied to the output of KV cache concatenation and moves them to the individual concat inputs instead.KVCacheConcattransformation, which identifies KV concatenation patterns where both cache and new KV data undergo identical quantization (downconvert) followed by dequantization (upconvert). It optimizes the pattern by:KVCacheStaticQuantizationtest class, which covers all the new introduced transformations as a single pipeline.Tickets: