| quant_stage: | |
| quant_modifiers: | |
| QuantizationModifier: | |
| config_groups: | |
| fp8_attention_qkv_proj: | |
| weights: {num_bits: 8, type: float, strategy: tensor} | |
| output_activations: {num_bits: 8, type: float, strategy: channel, dynamic: false, | |
| symmetric: true} | |
| targets: ['re:.*q_proj'] | |