nvfp4_moe-e2e / recipe.yaml
nm-autobot's picture
Upload folder using huggingface_hub
c46f29a verified
Raw
History Blame Contribute Delete
230 Bytes
default_stage:
default_modifiers:
QuantizationModifier:
targets: [Linear]
ignore: [lm_head, 're:.*mlp.gate[.].*']
scheme: NVFP4
bypass_divisibility_checks: false
requires_calibration_data: true