mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-02-16 07:53:55 +08:00
[None][fix] Add cacheSaltID property to BlockKey serialization code (#11457)
Signed-off-by: thorjohnsen <41591019+thorjohnsen@users.noreply.github.com>
This commit is contained in:
parent
2989bf5b39
commit
29e44dd749
@ -2519,6 +2519,7 @@ size_t Serialization::serializedSize(tensorrt_llm::batch_manager::kv_cache_manag
|
||||
totalSize += su::serializedSize(key.uniqueTokens);
|
||||
// std::vector<MmKey> where MmKey is pair<std::array<uint8_t,32>, SizeType32>
|
||||
totalSize += su::serializedSize(key.extraKeys);
|
||||
totalSize += su::serializedSize(key.cacheSaltID);
|
||||
return totalSize;
|
||||
}
|
||||
|
||||
@ -2528,6 +2529,7 @@ void Serialization::serialize(tensorrt_llm::batch_manager::kv_cache_manager::Blo
|
||||
su::serialize(key.loraTaskId, os);
|
||||
su::serialize(key.uniqueTokens, os);
|
||||
su::serialize(key.extraKeys, os);
|
||||
su::serialize(key.cacheSaltID, os);
|
||||
}
|
||||
|
||||
tensorrt_llm::batch_manager::kv_cache_manager::BlockKey Serialization::deserializeBlockKey(std::istream& is)
|
||||
@ -2536,11 +2538,13 @@ tensorrt_llm::batch_manager::kv_cache_manager::BlockKey Serialization::deseriali
|
||||
auto loraTaskId = su::deserialize<std::optional<tensorrt_llm::batch_manager::kv_cache_manager::LoraTaskIdType>>(is);
|
||||
auto uniqueTokens = su::deserialize<std::vector<tensorrt_llm::runtime::UniqueToken>>(is);
|
||||
auto extraKeys = su::deserialize<std::vector<tensorrt_llm::batch_manager::kv_cache_manager::MmKey>>(is);
|
||||
auto cacheSaltID = su::deserialize<std::optional<CacheSaltIDType>>(is);
|
||||
tensorrt_llm::batch_manager::kv_cache_manager::BlockKey key;
|
||||
key.usesExtraIds = usesExtraIds;
|
||||
key.loraTaskId = std::move(loraTaskId);
|
||||
key.uniqueTokens = std::move(uniqueTokens);
|
||||
key.extraKeys = std::move(extraKeys);
|
||||
key.cacheSaltID = std::move(cacheSaltID);
|
||||
return key;
|
||||
}
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user