|
{ |
|
"measurement": { |
|
"model.layers.0": { |
|
"accuracy": 0.9371232986450195, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.1": { |
|
"accuracy": 0.9321787357330322, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.2": { |
|
"accuracy": 0.9608855843544006, |
|
"total_bits": 456407296, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.02, |
|
0.98 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.02, |
|
0.98 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.3": { |
|
"accuracy": 0.9555550813674927, |
|
"total_bits": 456407296, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.02, |
|
0.98 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.01, |
|
0.99 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.02, |
|
0.98 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.4": { |
|
"accuracy": 0.9511148929595947, |
|
"total_bits": 469252352, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.5": { |
|
"accuracy": 0.9530089497566223, |
|
"total_bits": 513030400, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.4, |
|
0.6 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.4, |
|
0.6 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.6": { |
|
"accuracy": 0.9506196975708008, |
|
"total_bits": 513030400, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.4, |
|
0.6 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.4, |
|
0.6 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.7": { |
|
"accuracy": 0.9419316053390503, |
|
"total_bits": 513030400, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.4, |
|
0.6 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.05, |
|
0.95 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.4, |
|
0.6 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.8": { |
|
"accuracy": 0.9573614597320557, |
|
"total_bits": 607664384, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.1, |
|
0.9 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.1, |
|
0.9 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.1, |
|
0.9 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.1, |
|
0.9 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.1, |
|
0.9 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.9": { |
|
"accuracy": 0.9613507986068726, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.10": { |
|
"accuracy": 0.96047043800354, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.11": { |
|
"accuracy": 0.960319995880127, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.12": { |
|
"accuracy": 0.9560093879699707, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.13": { |
|
"accuracy": 0.9528203010559082, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.14": { |
|
"accuracy": 0.9506630897521973, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.15": { |
|
"accuracy": 0.9465978145599365, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.16": { |
|
"accuracy": 0.9447896480560303, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.17": { |
|
"accuracy": 0.9538173675537109, |
|
"total_bits": 724056320, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.18": { |
|
"accuracy": 0.953197717666626, |
|
"total_bits": 724056320, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.19": { |
|
"accuracy": 0.9511487483978271, |
|
"total_bits": 724056320, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.20": { |
|
"accuracy": 0.9419491291046143, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.21": { |
|
"accuracy": 0.9437572956085205, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.22": { |
|
"accuracy": 0.9443583488464355, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.23": { |
|
"accuracy": 0.9421937465667725, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.24": { |
|
"accuracy": 0.9423227310180664, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.25": { |
|
"accuracy": 0.9409959316253662, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.26": { |
|
"accuracy": 0.951786994934082, |
|
"total_bits": 724056320, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.5, |
|
0.5 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.27": { |
|
"accuracy": 0.9403626918792725, |
|
"total_bits": 646461696, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1.0 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128, |
|
"2": 128 |
|
}, |
|
"bits": [ |
|
4, |
|
2 |
|
], |
|
"bits_prop": [ |
|
0.25, |
|
0.75 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.28": { |
|
"accuracy": 0.9786742925643921, |
|
"total_bits": 879245568, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.29": { |
|
"accuracy": 0.9775421619415283, |
|
"total_bits": 879245568, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.30": { |
|
"accuracy": 0.9752198457717896, |
|
"total_bits": 879245568, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.31": { |
|
"accuracy": 0.9711686372756958, |
|
"total_bits": 879245568, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
} |
|
} |
|
} |