| { | |
| "measurement": { | |
| "model.layers.0": { | |
| "accuracy": 0.9522858953569084, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.1": { | |
| "accuracy": 0.9113649652572349, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.2": { | |
| "accuracy": 0.5754014626145363, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.3": { | |
| "accuracy": 0.8346537635661662, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.4": { | |
| "accuracy": 0.9821812690061051, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.5": { | |
| "accuracy": 0.9811006280651782, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.6": { | |
| "accuracy": 0.9804062172188424, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.7": { | |
| "accuracy": 0.9799818968458567, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.8": { | |
| "accuracy": 0.9785304359684233, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.9": { | |
| "accuracy": 0.9782906010514125, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.10": { | |
| "accuracy": 0.9774000000325032, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.11": { | |
| "accuracy": 0.9770228461129591, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.12": { | |
| "accuracy": 0.9764093302073888, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.13": { | |
| "accuracy": 0.9736090413061902, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.14": { | |
| "accuracy": 0.9728761716396548, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.15": { | |
| "accuracy": 0.9717501879786141, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.16": { | |
| "accuracy": 0.9719149245065637, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.17": { | |
| "accuracy": 0.971333933644928, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.18": { | |
| "accuracy": 0.9697856259881519, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.19": { | |
| "accuracy": 0.9718488412327133, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.20": { | |
| "accuracy": 0.9690313608734868, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.21": { | |
| "accuracy": 0.9688104883534834, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.22": { | |
| "accuracy": 0.9654940785258077, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.23": { | |
| "accuracy": 0.9653943074517883, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.24": { | |
| "accuracy": 0.9633830261882395, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.25": { | |
| "accuracy": 0.9626653084997088, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.26": { | |
| "accuracy": 0.9641250073327683, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.27": { | |
| "accuracy": 0.9639517902396619, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.28": { | |
| "accuracy": 0.965472717594821, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.29": { | |
| "accuracy": 0.9657354074879549, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.30": { | |
| "accuracy": 0.968929739901796, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.31": { | |
| "accuracy": 0.9708427549339831, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.32": { | |
| "accuracy": 0.9665285683004186, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.33": { | |
| "accuracy": 0.966284410096705, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.34": { | |
| "accuracy": 0.9638014331576414, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.35": { | |
| "accuracy": 0.9620210637804121, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.36": { | |
| "accuracy": 0.9612511987797916, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.37": { | |
| "accuracy": 0.9527971904026344, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.38": { | |
| "accuracy": 0.9518032884225249, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.39": { | |
| "accuracy": 0.947145561571233, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.40": { | |
| "accuracy": 0.9391240942059085, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.41": { | |
| "accuracy": 0.9377131020883098, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.42": { | |
| "accuracy": 0.9319567781640217, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.43": { | |
| "accuracy": 0.9371241622138768, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.44": { | |
| "accuracy": 0.9392979013500735, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.45": { | |
| "accuracy": 0.9402275587199256, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.46": { | |
| "accuracy": 0.9368646148359403, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.47": { | |
| "accuracy": 0.9256984132807702, | |
| "total_bits": 2541146112.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| } | |
| } | |
| } |