You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Somehow, we ended up with two quantization options for queries. First, we have --quantized, and then we have --scorer quantized. I think the easier way would be to remove --quantized because the other takes part in scorer resolution directly. The alternative is to have exactly one of the options required because saying --scorer bm25 --quantized makes no sense.
The text was updated successfully, but these errors were encountered:
It is not exactly redundant. --quantized is used by https://github.com/pisa-engine/pisa/blob/master/include/pisa/wand_data_compressed.hpp
If --quantized is passed, when block upper bound scores are computed they do not need to be unquantized. Please note that in https://github.com/pisa-engine/pisa/blob/master/include/pisa/wand_data_compressed.hpp scores are stored quantized.
We could find a way to pass it only once to the CLI, but we will need to pass it down to wand_data_compressed anyway.
I understand, but we can easily use auto quantized = scorer_name == "quantized" and pass it along. The point is to have the command arguments that make sense. I've been trying to run quantized queries with --quantized and it just spit out some garbage results because I didn't also say -s quantized. This is not a nuclear weapon, we don't need two triggers.
Somehow, we ended up with two quantization options for
queries
. First, we have--quantized
, and then we have--scorer quantized
. I think the easier way would be to remove--quantized
because the other takes part in scorer resolution directly. The alternative is to have exactly one of the options required because saying--scorer bm25 --quantized
makes no sense.The text was updated successfully, but these errors were encountered: