Supported Models
embedds architecture is rooted in the concept of a ModelClass - these represent a collection of fine tuned models that have a similar graph architecture and positioning in the market. Non-exhaustive list of available models is as follows:
| Model Class | Model Name | Size |
|---|---|---|
| ModelClass_E5 | E5_LARGE | 1.2GB |
| ModelClass_E5 | E5_BASE | 418MB |
| ModelClass_E5 | E5_SMALL | 128MB |
| ModelClass_INSTRUCTOR | INSTRUCTOR_LARGE | 1.5GB |
| ModelClass_INSTRUCTOR | INSTRUCTOR_BASE | 421MB |
| ModelClass_CLIP | RN50_OPENAI | textual: 255MB | visual: 153MB |
| ModelClass_CLIP | RN50_YFCC15M | textual: 255MB | visual: 153MB |
| ModelClass_CLIP | RN50_CC12M | textual: 255MB | visual: 153MB |
| ModelClass_CLIP | VIT_L_14_336_OPENAI | textual: 473MB | visual: 1.2GB |
Note: For more information about CLIP class models, please look at clip_models.rs.
Including a Fine Tuned Model
this is currently a WIP - see this GitHub issue for progress.