Benchmarking and Core Logic Updates #26
Conversation
|
@luis-gasparschroeder
|
| vector_db: VectorDB = HNSWLibVectorDB(), | ||
| embedding_metadata_storage: EmbeddingMetadataStorage = InMemoryEmbeddingMetadataStorage(), | ||
| eviction_policy: EvictionPolicy = LRUEvictionPolicy(), | ||
| eviction_policy: EvictionPolicy = NoEvictionPolicy(), |
There was a problem hiding this comment.
Are we going to have a NoEvictionPolicy?
What will happen if the cache is full
There was a problem hiding this comment.
As long as LRU is not implemented, let's have NoEviction. Otherwise, people expect things to be evicted
| ) | ||
|
|
||
| for i, delta in enumerate(vectorq_local_deltas): | ||
| for i, _ in enumerate(vectorq_local_latencies): |
There was a problem hiding this comment.
I think you can just use range(len(vectorq_local_latencies)) instead
There was a problem hiding this comment.
What is the main purpose of this change? (switching methods order)
There was a problem hiding this comment.
Why splitting the original is_global param to two different policies?
Do they have many differences? If >80% are the same, I think it's nicer to keep it with one file like the original one.
There was a problem hiding this comment.
We will likely adjust the likelihood computation for the global one. I agree that having both leads to duplicated code, but it's still learn than having if/else statements all over the place in one file.
2 participants
Implemented the core VectorQ policy logic.