[Feature] Prefix sharing. #53
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
Signed-off-by: Duyi-Wang <[email protected]>
pujiang2018
reviewed
Nov 23, 2023
src/models/kvcache_manager.h
Outdated
| @@ -24,27 +24,41 @@ class KVCacheManager { | |||
| this->layers = layers; | |||
| this->cachedKeys = new KVCacheTensor<KVCacheT>[layers]; | |||
| this->cachedValues = new KVCacheTensor<KVCacheT>[layers]; | |||
| this->cachedPrefixKeys = new KVCacheTensor<KVCacheT>[layers]; | |||
There was a problem hiding this comment.
If prefix_sharing=false, do not need to allocate it (although small memory).
Suggest allocating it when really needed.
| this->getPositionIds(prefixIDs, batchSize, pastSeqLen, 0); | ||
|
|
||
| free(prefixIDs); | ||
| ids = newIDs; |
There was a problem hiding this comment.
any chance to free the ID in future since it is dynamically allocated?
|
|
||
| this->prepareAttnMask(prefixIDs, 0); | ||
|
|
||
| this->getPositionIds(prefixIDs, batchSize, pastSeqLen, 0); |
There was a problem hiding this comment.
Do we really need to call getPositionIds?
| p[keyLen - 1] * ctx->attFactor); | ||
| p[2] * ctx->attFactor, p[strideC - 3] * ctx->attFactor, p[strideC - 2] * ctx->attFactor, | ||
| p[strideC - 1] * ctx->attFactor); | ||
| // for (int qki = 0; qki < queryLen; qki++) { |
There was a problem hiding this comment.
if not need, pls remove such commented code.
There was a problem hiding this comment.
This is used to print the whole QK score and attention mask matrix.
src/models/common_decoder.h
Outdated
| memcpy(newIDs + inputSeqLen * bs, ids + seqLen * bs + pastSeqLen, inputSeqLen * sizeof(int)); | ||
| } | ||
|
|
||
| this->prepareAttnMask(prefixIDs, 0); |
There was a problem hiding this comment.
The purpose of this step is?
pujiang2018
approved these changes
Nov 28, 2023
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.
Support Llama, chatGLM2, Baichuan, and Opt. Not support chatGLM 1 model.