| { |
| "measurement": { |
| "model.layers.0": { |
| "accuracy": 0.9678027320187539, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.1": { |
| "accuracy": 0.972175901173614, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.2": { |
| "accuracy": 0.9736290038563311, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.3": { |
| "accuracy": 0.9708406269201078, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.4": { |
| "accuracy": 0.9729515804210678, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.5": { |
| "accuracy": 0.9738428902928717, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.6": { |
| "accuracy": 0.9741117567173205, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.7": { |
| "accuracy": 0.9888297048019012, |
| "total_bits": 411586560.0, |
| "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": 0.9892435610818211, |
| "total_bits": 411586560.0, |
| "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": 0.988469355419511, |
| "total_bits": 411586560.0, |
| "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": 0.9884800493164221, |
| "total_bits": 411586560.0, |
| "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": 0.9891470322618261, |
| "total_bits": 411586560.0, |
| "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": 0.9886148972582305, |
| "total_bits": 411586560.0, |
| "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": 0.9881799372960813, |
| "total_bits": 411586560.0, |
| "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": 0.988099456357304, |
| "total_bits": 411586560.0, |
| "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": 0.9880573975678999, |
| "total_bits": 411586560.0, |
| "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": 0.9730851845233701, |
| "total_bits": 411586560.0, |
| "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": 0.9901754914753838, |
| "total_bits": 411586560.0, |
| "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": 0.9895372683531605, |
| "total_bits": 411586560.0, |
| "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": 0.9861618995782919, |
| "total_bits": 411586560.0, |
| "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": 0.9870588034391403, |
| "total_bits": 411586560.0, |
| "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": 0.9842094271443784, |
| "total_bits": 411586560.0, |
| "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": 0.9877904353779741, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.23": { |
| "accuracy": 0.9865320580720436, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.24": { |
| "accuracy": 0.9857799936435185, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.25": { |
| "accuracy": 0.981846509152092, |
| "total_bits": 411586560.0, |
| "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": 0.98228212093818, |
| "total_bits": 411586560.0, |
| "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": 0.9861528667097446, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.28": { |
| "accuracy": 0.9840018938702997, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.29": { |
| "accuracy": 0.985124388418626, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.30": { |
| "accuracy": 0.9838031140970998, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.31": { |
| "accuracy": 0.9842333355918527, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.32": { |
| "accuracy": 0.9819523809419479, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.33": { |
| "accuracy": 0.9800734139280394, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.34": { |
| "accuracy": 0.9797778406646103, |
| "total_bits": 435240960.0, |
| "o_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "down_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "q_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "k_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "v_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "gate_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| }, |
| "up_proj": { |
| "group_size": { |
| "4": 32 |
| }, |
| "bits": [ |
| 4 |
| ], |
| "bits_prop": [ |
| 1 |
| ], |
| "scale_bits": 4, |
| "scale_groups:": 32 |
| } |
| }, |
| "model.layers.35": { |
| "accuracy": 0.9364171591587365, |
| "total_bits": 411586560.0, |
| "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 |
| } |
| } |
| } |
| } |