-
Notifications
You must be signed in to change notification settings - Fork 148
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
LoRA Adapter #871
Draft
slyalin
wants to merge
35
commits into
openvinotoolkit:master
Choose a base branch
from
slyalin:dynamic_lora
base: master
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Draft
LoRA Adapter #871
+1,869
−311
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
… for ReadValue. It works around a bug in CPU with state initialization.
…nd separate base LoRA classes reuse outside (e.g. in SD). Code duplication, not compiling.
…elines with new API (+ several WAs for CPU bugs exposed during testing with various adapters).
…config for more convenient user experience.
…aptersConfig methods. Compilation is broken.
…tensors and non-empty tensors of minimal size.
…generation twice for each variant.
…e transformation activated by a new config flag.
…pha is not specified in command line
… defaults, less need of explicit use of AdapterConfig, introduced mode instead of bool flags.
…DE_STATIC explicitly in SD and lcm. Sorted examples in lora sample from static to dynamic. Passed MODE_STATIC_RANK to internal rank calculation.
…ependency. Switch lora sample to CPU. Disable DEBUG_PRINT in Release configuration.
… behaviour in the scenario when adapters are switched without setting alpha.
…ed cached alpha constants -- creating them each time, because they are small.
…apters simultaniously.
…a couple of TODOs: migrated to InferRequestCache. Reshafled the order of functions for a more fluent review.
…ngle class because there is no need for multiple classes in hierachy. Device-dependent LoRA mode deduction in AdapterController. Passed device type in all places where applicable.
…nd compile time as well as LoRA preparation time separately.
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.
TODO:
AnyMap
with properties, and aligned withgenerate
. A trivial attempt to do that failed due to a specific implementation ofProperty::Forward
helper class that confuses the compiler if multiple such arguments are passed. Requires a more accurate definition ofAdatperConfig
constructors.lora_greedy_causal_lm
sample instead of a series of different modes.