Add kv_cache size to pipeline generation log.#3989
Open
Conversation
mzegla
reviewed
Feb 18, 2026
Collaborator
mzegla
left a comment
mzegla
left a comment
There was a problem hiding this comment.
Does not seem to make sense to report 100% of X in every log for dynamic cache size.
It's a minor, but can we drop showing percentage for dynamic cache size and only report size in MB/GB etc. ?
### 🛠 Summary [CVS-180948](https://jira.devtools.intel.com/browse/CVS-180948) Adding new parameter to export_model to enable export of model with requirement for trusting remote code. ### 🧪 Checklist - [ ] Unit tests added. - [ ] The documentation updated. - [ ] Change follows security best practices. `` --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Miłosz Żeglarski <milosz.zeglarski@intel.com>
### 🛠 Summary Adding environmental variable for Qwen3-Coder-30B to make it work correctly ### 🧪 Checklist - [ ] Unit tests added. - [ ] The documentation updated. - [ ] Change follows security best practices. ``
### 🛠 Summary Updating export and NPU usage. ### 🧪 Checklist - [ ] Unit tests added. - [ ] The documentation updated. - [ ] Change follows security best practices. `` --------- Co-authored-by: Damian Kalinowski <damian.kalinowski@intel.com> Co-authored-by: Trawinski, Dariusz <dariusz.trawinski@intel.com>
mzegla
approved these changes
Mar 6, 2026
Contributor
There was a problem hiding this comment.
Pull request overview
Updates OVMS GenAI continuous batching logging to include kv-cache size (and whether the cache is treated as dynamic vs static), and bumps pinned OpenVINO/GenAI/tokenizers dependencies to versions that provide the required metrics field.
Changes:
- Extend continuous batching metrics log line to print cache usage along with kv-cache size (bytes formatted as B/KB/MB/GB/TB).
- Propagate a “dynamic kv-cache” flag into
LLMExecutorWrapperbased onschedulerConfig.cache_size == 0. - Bump OpenVINO / OpenVINO GenAI / OpenVINO tokenizers pinning (Makefile, Windows deps script, demo export requirements) to
dev20260305/ updated commit hashes.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
src/llm/language_model/continuous_batching/llm_executor.hpp |
Adds formatting helpers and logs kv-cache size + dynamic/static label in metrics output. |
src/llm/language_model/continuous_batching/servable_initializer.cpp |
Passes dynamic/static cache indicator into LLMExecutorWrapper. |
Makefile |
Updates pinned OpenVINO/GenAI/tokenizers branches and nightly package URLs. |
windows_install_build_dependencies.bat |
Updates default GenAI nightly ZIP URL and pinned source branches on Windows. |
demos/common/export_models/requirements.txt |
Updates OpenVINO and tokenizers Python package versions to dev20260305. |
Comment on lines
19
to
+67
| #include <atomic> | ||
| #include <condition_variable> | ||
| #include <cstdint> | ||
| #include <memory> | ||
| #include <mutex> | ||
| #include <string> | ||
| #include <thread> | ||
| #include <utility> | ||
|
|
||
| #include <openvino/genai/continuous_batching_pipeline.hpp> | ||
|
|
||
| #include "../../../logging.hpp" | ||
| #include "../../../profiler.hpp" | ||
|
|
||
| namespace ovms { | ||
| struct LLMExecutor { | ||
| bool isDynamicKVCache; | ||
| // For logging purposes we could have more information about graph and node here | ||
| std::mutex mutex; | ||
| std::condition_variable cv; | ||
| std::shared_ptr<ov::genai::ContinuousBatchingPipeline> pipe = nullptr; | ||
|
|
||
| LLMExecutor(std::shared_ptr<ov::genai::ContinuousBatchingPipeline> pipe) { | ||
| LLMExecutor(std::shared_ptr<ov::genai::ContinuousBatchingPipeline> pipe, bool isDynamicKVCacheSet = false) { | ||
| this->pipe = std::move(pipe); | ||
| this->isDynamicKVCache = isDynamicKVCacheSet; | ||
| } | ||
|
|
||
| bool hasRequests() { | ||
| return (pipe->has_non_finished_requests()); | ||
| } | ||
|
|
||
| void step() { | ||
| OVMS_PROFILE_FUNCTION(); | ||
| pipe->step(); | ||
| } | ||
|
|
||
| void waitForRequests(std::atomic<bool>* receivedEndSignal) { | ||
| std::unique_lock<std::mutex> lock(mutex); | ||
| cv.wait(lock, [this, receivedEndSignal] { return (pipe->has_non_finished_requests() || *receivedEndSignal); }); | ||
| } | ||
|
|
||
| void notify() { | ||
| std::unique_lock<std::mutex> lock(mutex); | ||
| cv.notify_one(); | ||
| } | ||
|
|
||
| std::string formatCacheInfo(float cacheUsage, size_t cacheBytes, bool isCacheDynamic) { | ||
| std::ostringstream oss; | ||
| oss << std::fixed << std::setprecision(1); |
There was a problem hiding this comment.
This header now uses std::ostringstream and i/o manipulators (std::fixed / std::setprecision) but does not include the required standard headers ( and ). This will fail to compile in TUs that don't already include them indirectly; add the missing includes here (and consider for size_t to keep the header self-contained).
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
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
JIRA CVS-181044
Depends on https://github.com/openvinotoolkit/openvino.genai/pull/3352/changes and ovms genai update.
Output:
Cache usage static 18.6% of 1022.4 MB;
for static 1GB
Cache usage dynamic 98.6% of 189.9 MB;
for dynamic cache_size = 0
For: {"prompt_tokens":2299,"completion_tokens":30,"total_tokens":2329}}
🧪 Checklist
``