{ "summary": { "group_size": 64, "tensors_ternary": 0, "tensors_int8": 99, "tensors_total": 156, "tensors_quantized": 99, "float_mb": 100.72, "result_mb": 26.79, "ratio": 3.76, "worst_relative_error": 0.0063, "mean_relative_error": 0.006, "mean_zero_fraction": 0.0084, "method": "int8", "quality": { "tokens_compared": 95, "top1_agreement": 0.9684, "top5_agreement": 1.0, "mean_kl_divergence": 0.0017, "reference_perplexity": 49.146, "candidate_perplexity": 49.537, "perplexity_ratio": 1.008, "greedy_identical_prefix": [ 12, 24, 24, 24, 24 ], "greedy_identical_answers": "7/8" } }, "tensors": [ { "name": "transformer.h.0.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.0.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00583, "zero_fraction": 0.00865 }, { "name": "transformer.h.0.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00602, "zero_fraction": 0.00944 }, { "name": "transformer.h.0.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.0.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00591, "zero_fraction": 0.00816 }, { "name": "transformer.h.0.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00605, "zero_fraction": 0.00846 }, { "name": "transformer.h.0.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.0.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.0.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00603, "zero_fraction": 0.00841 }, { "name": "transformer.h.0.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00816 }, { "name": "transformer.h.0.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00822 }, { "name": "transformer.h.1.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.1.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00599, "zero_fraction": 0.00838 }, { "name": "transformer.h.1.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00844 }, { "name": "transformer.h.1.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.1.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00604, "zero_fraction": 0.00813 }, { "name": "transformer.h.1.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00897 }, { "name": "transformer.h.1.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.1.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.1.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.006, "zero_fraction": 0.00813 }, { "name": "transformer.h.1.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.00814 }, { "name": "transformer.h.1.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00805 }, { "name": "transformer.h.10.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.10.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00607, "zero_fraction": 0.00848 }, { "name": "transformer.h.10.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00592, "zero_fraction": 0.00844 }, { "name": "transformer.h.10.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.10.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00594, "zero_fraction": 0.00785 }, { "name": "transformer.h.10.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00605, "zero_fraction": 0.00832 }, { "name": "transformer.h.10.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.10.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.10.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.006, "zero_fraction": 0.00855 }, { "name": "transformer.h.10.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00592, "zero_fraction": 0.00817 }, { "name": "transformer.h.10.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00594, "zero_fraction": 0.00822 }, { "name": "transformer.h.11.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.11.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00626, "zero_fraction": 0.00885 }, { "name": "transformer.h.11.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00777 }, { "name": "transformer.h.11.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.11.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.0084 }, { "name": "transformer.h.11.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00791 }, { "name": "transformer.h.11.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.11.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.11.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00601, "zero_fraction": 0.0082 }, { "name": "transformer.h.11.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00593, "zero_fraction": 0.00824 }, { "name": "transformer.h.11.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.00834 }, { "name": "transformer.h.12.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.12.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00615, "zero_fraction": 0.0084 }, { "name": "transformer.h.12.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00601, "zero_fraction": 0.00823 }, { "name": "transformer.h.12.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.12.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00864 }, { "name": "transformer.h.12.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.006, "zero_fraction": 0.00781 }, { "name": "transformer.h.12.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.12.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.12.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00601, "zero_fraction": 0.00821 }, { "name": "transformer.h.12.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00817 }, { "name": "transformer.h.12.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00801 }, { "name": "transformer.h.13.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.13.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00599, "zero_fraction": 0.00773 }, { "name": "transformer.h.13.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00606, "zero_fraction": 0.00863 }, { "name": "transformer.h.13.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.13.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00827 }, { "name": "transformer.h.13.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00606, "zero_fraction": 0.00928 }, { "name": "transformer.h.13.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.13.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.13.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00614, "zero_fraction": 0.00856 }, { "name": "transformer.h.13.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00592, "zero_fraction": 0.00835 }, { "name": "transformer.h.13.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00825 }, { "name": "transformer.h.2.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.2.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00601, "zero_fraction": 0.00887 }, { "name": "transformer.h.2.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00812 }, { "name": "transformer.h.2.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.2.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.00829 }, { "name": "transformer.h.2.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00893 }, { "name": "transformer.h.2.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.2.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.2.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00798 }, { "name": "transformer.h.2.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.00834 }, { "name": "transformer.h.2.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00831 }, { "name": "transformer.h.3.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.3.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00608, "zero_fraction": 0.00834 }, { "name": "transformer.h.3.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00592, "zero_fraction": 0.00808 }, { "name": "transformer.h.3.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.3.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00604, "zero_fraction": 0.00793 }, { "name": "transformer.h.3.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00602, "zero_fraction": 0.00958 }, { "name": "transformer.h.3.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.3.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.3.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00831 }, { "name": "transformer.h.3.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.00805 }, { "name": "transformer.h.3.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00792 }, { "name": "transformer.h.4.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.4.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00593, "zero_fraction": 0.00832 }, { "name": "transformer.h.4.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00853 }, { "name": "transformer.h.4.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.4.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00835 }, { "name": "transformer.h.4.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00594, "zero_fraction": 0.0084 }, { "name": "transformer.h.4.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.4.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.4.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00822 }, { "name": "transformer.h.4.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00856 }, { "name": "transformer.h.4.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00826 }, { "name": "transformer.h.5.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.5.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00608, "zero_fraction": 0.0083 }, { "name": "transformer.h.5.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00593, "zero_fraction": 0.0081 }, { "name": "transformer.h.5.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.5.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00593, "zero_fraction": 0.00823 }, { "name": "transformer.h.5.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00599, "zero_fraction": 0.00869 }, { "name": "transformer.h.5.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.5.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.5.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00602, "zero_fraction": 0.00835 }, { "name": "transformer.h.5.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00602, "zero_fraction": 0.00861 }, { "name": "transformer.h.5.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00599, "zero_fraction": 0.00843 }, { "name": "transformer.h.6.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.6.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00607, "zero_fraction": 0.00883 }, { "name": "transformer.h.6.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00602, "zero_fraction": 0.00898 }, { "name": "transformer.h.6.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.6.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00821 }, { "name": "transformer.h.6.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00603, "zero_fraction": 0.00903 }, { "name": "transformer.h.6.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.6.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.6.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00608, "zero_fraction": 0.00835 }, { "name": "transformer.h.6.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00808 }, { "name": "transformer.h.6.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00831 }, { "name": "transformer.h.7.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.7.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00628, "zero_fraction": 0.00859 }, { "name": "transformer.h.7.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00591, "zero_fraction": 0.00831 }, { "name": "transformer.h.7.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.7.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00603, "zero_fraction": 0.00844 }, { "name": "transformer.h.7.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00617, "zero_fraction": 0.00897 }, { "name": "transformer.h.7.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.7.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.7.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00604, "zero_fraction": 0.00819 }, { "name": "transformer.h.7.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00598, "zero_fraction": 0.00854 }, { "name": "transformer.h.7.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00831 }, { "name": "transformer.h.8.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.8.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00603, "zero_fraction": 0.00773 }, { "name": "transformer.h.8.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00599, "zero_fraction": 0.00867 }, { "name": "transformer.h.8.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.8.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00593, "zero_fraction": 0.00841 }, { "name": "transformer.h.8.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00609, "zero_fraction": 0.0095 }, { "name": "transformer.h.8.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.8.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.8.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.006, "zero_fraction": 0.00848 }, { "name": "transformer.h.8.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00595, "zero_fraction": 0.00816 }, { "name": "transformer.h.8.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00594, "zero_fraction": 0.0081 }, { "name": "transformer.h.9.attn.k_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.9.attn.k_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00618, "zero_fraction": 0.0084 }, { "name": "transformer.h.9.attn.o_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00597, "zero_fraction": 0.00798 }, { "name": "transformer.h.9.attn.q_norm.weight", "shape": [ 64 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.9.attn.q_proj.weight", "shape": [ 384, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00596, "zero_fraction": 0.00831 }, { "name": "transformer.h.9.attn.v_proj.weight", "shape": [ 128, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00606, "zero_fraction": 0.0096 }, { "name": "transformer.h.9.ln_1.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.9.ln_2.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.h.9.mlp.w_down.weight", "shape": [ 384, 1024 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00602, "zero_fraction": 0.00851 }, { "name": "transformer.h.9.mlp.w_gate.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00594, "zero_fraction": 0.00808 }, { "name": "transformer.h.9.mlp.w_up.weight", "shape": [ 1024, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00594, "zero_fraction": 0.00827 }, { "name": "transformer.ln_f.weight", "shape": [ 384 ], "quantized": false, "reason": "not a matrix", "relative_error": 0.0, "zero_fraction": 0.0 }, { "name": "transformer.wte.weight", "shape": [ 8192, 384 ], "quantized": true, "reason": "8-bit", "relative_error": 0.00606, "zero_fraction": 0.00844 } ] }