| default_stage: | |
| default_modifiers: | |
| QuantizationModifier: | |
| targets: [Linear] | |
| ignore: ['re:.*mlp\.router\.gate.*', lm_head, 're:.*norm.*', 're:.*self_attn.*'] | |
| scheme: FP8_BLOCK | |
| bypass_divisibility_checks: false | |
| default_stage: | |
| default_modifiers: | |
| QuantizationModifier: | |
| targets: [Linear] | |
| ignore: ['re:.*mlp\.router\.gate.*', lm_head, 're:.*norm.*', 're:.*self_attn.*'] | |
| scheme: FP8_BLOCK | |
| bypass_divisibility_checks: false | |