Skip to content

Implement quantization on fp16 directly without inflating to fp32 #16533

Description

@Pulkitg64

Description

Follow-up to #16473, which added scalar quantization support for the float16 vector encoding.

The quantization works on fp32, so every fp16 vector is inflated to a float[] before it is quantized, and the centroid and corrective terms are computed in fp32 as well. Since Java has no fp16 arithmetic type, so there is no way to compute in fp16 at the moment. Once Java have support for doing arithmetic operations on fp16 directly, we should switch our existing implementation.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions