Proposal for Adding Quantization to AIR

[gartia]

Proposal for Adding Quantization to AIR

I propose integrating a quantization field into the AIR schema to enhance its applicability beyond just stable diffusion models, extending its use to general AI applications. While quantization is not commonly associated with stable diffusion, incorporating quantized models can yield more unified naming schema.

Rationale:

Quantization methods vary widely, and when developing tools for inferencing, it is crucial to identify the file type directly. This approach avoids the need to parse the string for quantization details after determining the file type, streamlining the process.

Example:

GGML's SD inference demonstrates that quantization does not significantly degrade quality, especially considering the performance gains.

Proposed Regex Update:

To accommodate this change, I suggest updating our regex pattern as follows:

Grouped

^(?:urn:)?(?:air:)?(?:(?<ecosystem>[a-zA-Z0-9_\-\/]+):)?(?:(?<type>[a-zA-Z0-9_\-\/]+):)?(?<source>[a-zA-Z0-9_\-\/]+):(?<id>[a-zA-Z0-9_\-\/]+)(?:@(?<version>[a-zA-Z0-9_\-]+))?(?:\.(?<format>[a-zA-Z0-9_\-]+))?(?:\.(?<quantization>[a-zA-Z0-9_\-]+))?$

NonGrouped

^(?:urn:)?(?:air:)?(?:([a-zA-Z0-9_\-\/]+):)?(?:([a-zA-Z0-9_\-\/]+):)?([a-zA-Z0-9_\-\/]+):([a-zA-Z0-9_\-\/]+)(?:@([a-zA-Z0-9_\-]+))?(?:\.([a-zA-Z0-9_\-]+))?(?:\.([a-zA-Z0-9_\-]+))?$

Potential Issue:

The proposed regex update introduces a new format that is not recognized by the original regex pattern. This discrepancy could lead to compatibility issues where systems using the old regex pattern are unable to parse identifiers that include the new quantization field.