|
{ |
|
"measurement": { |
|
"model.layers.0": { |
|
"accuracy": 0.9235353469848633, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.1": { |
|
"accuracy": 0.7686681747436523, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.2": { |
|
"accuracy": 0.8663358688354492, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.3": { |
|
"accuracy": 0.8417501449584961, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.4": { |
|
"accuracy": 0.8675830364227295, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.5": { |
|
"accuracy": 0.8590667247772217, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.6": { |
|
"accuracy": 0.8517341613769531, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.7": { |
|
"accuracy": 0.8489260673522949, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.8": { |
|
"accuracy": 0.9279017448425293, |
|
"total_bits": 449761824, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.9": { |
|
"accuracy": 0.9575139284133911, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.10": { |
|
"accuracy": 0.95622718334198, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.11": { |
|
"accuracy": 0.949416995048523, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.12": { |
|
"accuracy": 0.9479968547821045, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.13": { |
|
"accuracy": 0.9440830945968628, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.14": { |
|
"accuracy": 0.9452991485595703, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.15": { |
|
"accuracy": 0.9413189888000488, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.16": { |
|
"accuracy": 0.944807767868042, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.17": { |
|
"accuracy": 0.9402191638946533, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.18": { |
|
"accuracy": 0.9394395351409912, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.19": { |
|
"accuracy": 0.9472098350524902, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.20": { |
|
"accuracy": 0.9527584314346313, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.21": { |
|
"accuracy": 0.9584027528762817, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.22": { |
|
"accuracy": 0.9615957736968994, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.23": { |
|
"accuracy": 0.9349331855773926, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.24": { |
|
"accuracy": 0.963486909866333, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.25": { |
|
"accuracy": 0.9610995054244995, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.26": { |
|
"accuracy": 0.9591997861862183, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.27": { |
|
"accuracy": 0.9536696672439575, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.28": { |
|
"accuracy": 0.9514199495315552, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.29": { |
|
"accuracy": 0.944045901298523, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.30": { |
|
"accuracy": 0.9456077814102173, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.31": { |
|
"accuracy": 0.9442343711853027, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.32": { |
|
"accuracy": 0.9421529769897461, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.33": { |
|
"accuracy": 0.9256293773651123, |
|
"total_bits": 360997440, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.34": { |
|
"accuracy": 0.9283227920532227, |
|
"total_bits": 365124672, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.35": { |
|
"accuracy": 0.9434576034545898, |
|
"total_bits": 449761824, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"v_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"o_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"up_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.36": { |
|
"accuracy": 0.9467594623565674, |
|
"total_bits": 486917664, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.37": { |
|
"accuracy": 0.9436817169189453, |
|
"total_bits": 486917664, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"gate_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.38": { |
|
"accuracy": 0.9477231502532959, |
|
"total_bits": 571557408, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.39": { |
|
"accuracy": 0.9525341987609863, |
|
"total_bits": 604586016, |
|
"q_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"k_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
}, |
|
"down_proj": { |
|
"group_size": { |
|
"4": 128 |
|
}, |
|
"bits": [ |
|
4 |
|
], |
|
"bits_prop": [ |
|
1 |
|
], |
|
"scale_bits": 4 |
|
} |
|
}, |
|
"model.layers.40": { |
|
"accuracy": 0.9732036590576172, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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.41": { |
|
"accuracy": 0.9750280380249023, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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.42": { |
|
"accuracy": 0.9755113124847412, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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.43": { |
|
"accuracy": 0.9758803844451904, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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.44": { |
|
"accuracy": 0.9759896993637085, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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.45": { |
|
"accuracy": 0.9787205457687378, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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.46": { |
|
"accuracy": 0.9827241897583008, |
|
"total_bits": 697480224, |
|
"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.47": { |
|
"accuracy": 0.9262123107910156, |
|
"total_bits": 664451616, |
|
"q_proj": { |
|
"group_size": { |
|
"2": 64 |
|
}, |
|
"bits": [ |
|
2 |
|
], |
|
"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 |
|
} |
|
} |
|
} |
|
} |