| default_stage: | |
| default_modifiers: | |
| QuantizationModifier: | |
| config_groups: | |
| group_nvfp4: | |
| targets: [Linear] | |
| weights: | |
| num_bits: 4 | |
| type: float | |
| symmetric: true | |
| group_size: 16 | |
| strategy: tensor_group | |
| block_structure: null | |
| dynamic: false | |
| actorder: null | |
| scale_dtype: torch.float8_e4m3fn | |
| zp_dtype: null | |
| observer: memoryless_minmax | |
| observer_kwargs: {} | |
| input_activations: | |
| num_bits: 4 | |
| type: float | |
| symmetric: true | |
| group_size: 16 | |
| strategy: tensor_group | |
| block_structure: null | |
| dynamic: local | |
| actorder: null | |
| scale_dtype: torch.float8_e4m3fn | |
| zp_dtype: null | |
| observer: static_minmax | |
| observer_kwargs: {} | |
| output_activations: null | |
| format: null | |
| targets: [Linear] | |
| ignore: ['re:.*mlp.gate$', 're:.*shared_expert_gate$', 're:.*embed_tokens$', 're:mtp\.layers\.\d+\.mlp\.gate$', | |
| 're:mtp\.layers\.\d+\.mlp\.shared_expert_gate$', 're:visual.*', 're:model.visual.*'] | |
| bypass_divisibility_checks: false | |