Quantization for program-data separation #15419
Description
🚀 The feature, motivation and pitch
Allow weight-sharing with quantized weights.
- Quantize foundation weights only (for lora case)
- Preserve weight names (general case)
Alternatives
No response
Additional context
No response
RFC (Optional)
No response
Metadata
Metadata
Assignees
Labels
No labels
Type
Projects
Status