clef-flash-NVFP4 / recipe.yaml
aleksblago's picture
Add clef-flash-NVFP4
261aa6c verified
Raw History Blame Contribute Delete
287 Bytes
default_stage:
default_modifiers:
QuantizationModifier:
targets: [Linear]
ignore: [lm_head, 're:.*visual.*', 're:.*linear_attn.in_proj_a$', 're:.*linear_attn.in_proj_b$']
scheme: NVFP4
bypass_divisibility_checks: false
requires_calibration_data: true