| { |
| "bits": 3, |
| "dynamic": { |
| "^model\\.language_model\\.layers\\.0\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.0\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.0\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.0\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.0\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.0\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.1\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.1\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.1\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.1\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.1\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.1\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.10\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.10\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.10\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.10\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.10\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.10\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.self_attn\\.o_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.11\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.12\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.12\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.12\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.12\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.12\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.12\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.13\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.13\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.13\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.13\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.13\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.13\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.14\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.14\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.14\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.14\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.14\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.14\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.self_attn\\.o_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.15\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.16\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.16\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.16\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.16\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.16\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.16\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.17\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.17\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.17\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.17\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.17\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.17\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.18\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.18\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.18\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.18\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.18\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.18\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.self_attn\\.o_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.19\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.2\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.2\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.2\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.2\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.2\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.2\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.20\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.20\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.20\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.20\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.20\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.20\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.21\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.21\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.21\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.21\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.21\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.21\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.22\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.22\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.22\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.22\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.22\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.22\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.self_attn\\.o_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.23\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.24\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.24\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.24\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.24\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.24\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.24\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.25\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.25\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.25\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.25\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.25\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.25\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.26\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.26\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.26\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.26\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.26\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.26\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.self_attn\\.o_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.27\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.28\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.28\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.28\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.28\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.28\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.28\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.29\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.29\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.29\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.29\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.29\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.29\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.self_attn\\.o_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.3\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.30\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.30\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.30\\.linear_attn\\.out_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.30\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.30\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.30\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.self_attn\\.o_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.31\\.self_attn\\.v_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.4\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.4\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.4\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.4\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.4\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.4\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.5\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.5\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.5\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.5\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.5\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.5\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.6\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.6\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.6\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.6\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.6\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.6\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.self_attn\\.k_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.self_attn\\.o_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.self_attn\\.q_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.7\\.self_attn\\.v_proj$": { |
| "bits": 4, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.8\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.8\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.8\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 64, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.8\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.8\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.8\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.9\\.linear_attn\\.in_proj_qkv$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.9\\.linear_attn\\.in_proj_z$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.9\\.linear_attn\\.out_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.9\\.mlp\\.down_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.9\\.mlp\\.gate_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| }, |
| "^model\\.language_model\\.layers\\.9\\.mlp\\.up_proj$": { |
| "bits": 3, |
| "group_size": 128, |
| "desc_act": false, |
| "act_group_aware": true, |
| "sym": true |
| } |
| }, |
| "group_size": 128, |
| "desc_act": false, |
| "lm_head": false, |
| "method": "gptq", |
| "quant_method": "gptq", |
| "format": "gptq", |
| "checkpoint_format": "gptq", |
| "pack_dtype": "int32", |
| "meta": { |
| "recipe": "SAGE-M40-v2", |
| "quantizer": [ |
| "gptqmodel:7.5.0" |
| ], |
| "uri": "https://github.com/modelcloud/gptqmodel", |
| "damp_percent": 0.05, |
| "damp_auto_increment": 0.01, |
| "static_groups": false, |
| "true_sequential": true, |
| "mse": 0.0, |
| "gptaq": null, |
| "foem": null, |
| "act_group_aware": true, |
| "fallback": { |
| "strategy": "rtn", |
| "threshold": "0.5%", |
| "smooth": null |
| }, |
| "offload_to_disk": true, |
| "telemetry": { |
| "device": false |
| }, |
| "offload_to_disk_path": "/tmp/gptqmodel_gjqvai9c", |
| "pack_impl": "gpu", |
| "gc_mode": "interval", |
| "wait_for_submodule_finalizers": false, |
| "auto_forward_data_parallel": true, |
| "dense_vram_strategy": "exclusive", |
| "dense_vram_strategy_devices": null, |
| "moe_vram_strategy": "exclusive", |
| "moe_vram_strategy_devices": null, |
| "mock_quantization": false, |
| "hessian": { |
| "chunk_size": null, |
| "chunk_bytes": null, |
| "staging_dtype": "float32", |
| "dedup_shared_inputs": true |
| } |
| }, |
| "sym": true |
| } |
