| { |
| "measurement": { |
| "model.layers.0": { |
| "accuracy": 0.9470930099487305, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.1": { |
| "accuracy": 0.44930458068847656, |
| "total_bits": 306241536.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "2": 64 |
| }, |
| "bits": [ |
| 2 |
| ], |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.2": { |
| "accuracy": 0.9330713748931885, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.3": { |
| "accuracy": 0.970420241355896, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.4": { |
| "accuracy": 0.9747236967086792, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.5": { |
| "accuracy": 0.9633625745773315, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.6": { |
| "accuracy": 0.9738354682922363, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.7": { |
| "accuracy": 0.9770342111587524, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.8": { |
| "accuracy": 0.9778237342834473, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.9": { |
| "accuracy": 0.9768979549407959, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.10": { |
| "accuracy": 0.9737787246704102, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.11": { |
| "accuracy": 0.9728798866271973, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.12": { |
| "accuracy": 0.9736931324005127, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.13": { |
| "accuracy": 0.9764463901519775, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.14": { |
| "accuracy": 0.9773211479187012, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.15": { |
| "accuracy": 0.9774775505065918, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.16": { |
| "accuracy": 0.9796525239944458, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.17": { |
| "accuracy": 0.9784044027328491, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.18": { |
| "accuracy": 0.9794589281082153, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.19": { |
| "accuracy": 0.979212760925293, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.20": { |
| "accuracy": 0.9766635894775391, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.21": { |
| "accuracy": 0.9790831208229065, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.22": { |
| "accuracy": 0.9801523685455322, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.23": { |
| "accuracy": 0.9790130853652954, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.24": { |
| "accuracy": 0.9794877767562866, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.25": { |
| "accuracy": 0.9761180877685547, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.26": { |
| "accuracy": 0.9733923673629761, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.27": { |
| "accuracy": 0.9720484018325806, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.28": { |
| "accuracy": 0.9757145643234253, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.29": { |
| "accuracy": 0.975440263748169, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.30": { |
| "accuracy": 0.9650356769561768, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.31": { |
| "accuracy": 0.9708733558654785, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.32": { |
| "accuracy": 0.9723145961761475, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.33": { |
| "accuracy": 0.9625141620635986, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.34": { |
| "accuracy": 0.9762517213821411, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.35": { |
| "accuracy": 0.9684627056121826, |
| "total_bits": 314302464.0, |
| "o_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "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 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 128 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| } |
| } |
| } |