| quant_stage: | |
| quant_modifiers: | |
| GPTQModifier: | |
| sequential_update: true | |
| dampening_frac: 0.1 | |
| ignore: [lm_head, 're:.*block_sparse_moe.gate'] | |
| config_groups: | |
| group_0: | |
| weights: {num_bits: 4, type: int, symmetric: true, strategy: group, group_size: 128, | |
| actorder: false, observer: mse} | |
| targets: [Linear] |