File size: 569 Bytes
c0330b4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | default_stage:
default_modifiers:
AWQModifier:
targets: [Linear]
ignore: [lm_head, 're:.*mlp.gate$', 're:.*mlp.shared_expert_gate$']
scheme: W4A16
mappings:
- smooth_layer: re:.*input_layernorm$
balance_layers: ['re:.*self_attn.q_proj$', 're:.*self_attn.k_proj$', 're:.*self_attn.v_proj$',
're:.*mlp.gate_proj$', 're:.*mlp.up_proj$']
- smooth_layer: re:.*v_proj$
balance_layers: ['re:.*o_proj$']
- smooth_layer: re:.*up_proj$
balance_layers: ['re:.*down_proj$']
duo_scaling: false
|