BananaMind-2-Pro-Preview-Chat-mixed / quantization_report.json
fontlab's picture
Add a quantized checkpoint
f52d75e verified
Raw
History Blame Contribute Delete
51.8 kB
{
"summary": {
"group_size": 64,
"tensors_ternary": 12,
"tensors_int8": 157,
"tensors_total": 266,
"tensors_quantized": 169,
"float_mb": 555.89,
"result_mb": 144.94,
"ratio": 3.84,
"worst_relative_error": 0.5833,
"mean_relative_error": 0.0436,
"mean_zero_fraction": 0.0397,
"method": "mixed",
"quality": {
"tokens_compared": 95,
"top1_agreement": 0.9053,
"top5_agreement": 0.9895,
"mean_kl_divergence": 0.0111,
"reference_perplexity": 33.305,
"candidate_perplexity": 33.282,
"perplexity_ratio": 0.999,
"greedy_identical_prefix": [
24,
6,
15,
24,
8
],
"greedy_identical_answers": "3/8"
}
},
"tensors": [
{
"name": "transformer.h.0.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.0.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.52544,
"zero_fraction": 0.44605
},
{
"name": "transformer.h.0.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.58326,
"zero_fraction": 0.45928
},
{
"name": "transformer.h.0.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.0.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.54151,
"zero_fraction": 0.44389
},
{
"name": "transformer.h.0.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00605,
"zero_fraction": 0.00874
},
{
"name": "transformer.h.0.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.0.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.0.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00615,
"zero_fraction": 0.00923
},
{
"name": "transformer.h.0.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00839
},
{
"name": "transformer.h.0.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00866
},
{
"name": "transformer.h.1.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.1.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00836
},
{
"name": "transformer.h.1.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00803
},
{
"name": "transformer.h.1.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.1.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00609,
"zero_fraction": 0.00812
},
{
"name": "transformer.h.1.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00851
},
{
"name": "transformer.h.1.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.1.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.1.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00837
},
{
"name": "transformer.h.1.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00813
},
{
"name": "transformer.h.1.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00842
},
{
"name": "transformer.h.10.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.10.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00881
},
{
"name": "transformer.h.10.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00811
},
{
"name": "transformer.h.10.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.10.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00815
},
{
"name": "transformer.h.10.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00863
},
{
"name": "transformer.h.10.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.10.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.10.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00606,
"zero_fraction": 0.00844
},
{
"name": "transformer.h.10.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00842
},
{
"name": "transformer.h.10.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00842
},
{
"name": "transformer.h.11.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.11.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00609,
"zero_fraction": 0.00852
},
{
"name": "transformer.h.11.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00816
},
{
"name": "transformer.h.11.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.11.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00835
},
{
"name": "transformer.h.11.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00893
},
{
"name": "transformer.h.11.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.11.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.11.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00833
},
{
"name": "transformer.h.11.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00827
},
{
"name": "transformer.h.11.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00838
},
{
"name": "transformer.h.12.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.12.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00623,
"zero_fraction": 0.00854
},
{
"name": "transformer.h.12.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00595,
"zero_fraction": 0.00832
},
{
"name": "transformer.h.12.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.12.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00605,
"zero_fraction": 0.00841
},
{
"name": "transformer.h.12.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00877
},
{
"name": "transformer.h.12.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.12.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.12.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00831
},
{
"name": "transformer.h.12.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00823
},
{
"name": "transformer.h.12.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00804
},
{
"name": "transformer.h.13.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.13.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00633,
"zero_fraction": 0.0086
},
{
"name": "transformer.h.13.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00595,
"zero_fraction": 0.00842
},
{
"name": "transformer.h.13.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.13.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00607,
"zero_fraction": 0.00852
},
{
"name": "transformer.h.13.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00878
},
{
"name": "transformer.h.13.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.13.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.13.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00842
},
{
"name": "transformer.h.13.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.0083
},
{
"name": "transformer.h.13.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00845
},
{
"name": "transformer.h.14.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.14.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00617,
"zero_fraction": 0.00831
},
{
"name": "transformer.h.14.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00816
},
{
"name": "transformer.h.14.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.14.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00824
},
{
"name": "transformer.h.14.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00866
},
{
"name": "transformer.h.14.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.14.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.14.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00833
},
{
"name": "transformer.h.14.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00824
},
{
"name": "transformer.h.14.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00838
},
{
"name": "transformer.h.15.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.15.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00878
},
{
"name": "transformer.h.15.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00827
},
{
"name": "transformer.h.15.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.15.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00831
},
{
"name": "transformer.h.15.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00865
},
{
"name": "transformer.h.15.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.15.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.15.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00607,
"zero_fraction": 0.0085
},
{
"name": "transformer.h.15.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00607,
"zero_fraction": 0.00849
},
{
"name": "transformer.h.15.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00606,
"zero_fraction": 0.00852
},
{
"name": "transformer.h.16.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.16.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00615,
"zero_fraction": 0.00891
},
{
"name": "transformer.h.16.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00821
},
{
"name": "transformer.h.16.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.16.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00843
},
{
"name": "transformer.h.16.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00904
},
{
"name": "transformer.h.16.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.16.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.16.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00612,
"zero_fraction": 0.0085
},
{
"name": "transformer.h.16.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00605,
"zero_fraction": 0.00848
},
{
"name": "transformer.h.16.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.17.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.17.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0064,
"zero_fraction": 0.00907
},
{
"name": "transformer.h.17.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00821
},
{
"name": "transformer.h.17.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.17.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00619,
"zero_fraction": 0.0089
},
{
"name": "transformer.h.17.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00653,
"zero_fraction": 0.01005
},
{
"name": "transformer.h.17.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.17.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.17.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00826
},
{
"name": "transformer.h.17.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00842
},
{
"name": "transformer.h.17.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.00832
},
{
"name": "transformer.h.18.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.18.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00616,
"zero_fraction": 0.00827
},
{
"name": "transformer.h.18.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00822
},
{
"name": "transformer.h.18.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.18.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.0087
},
{
"name": "transformer.h.18.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00608,
"zero_fraction": 0.00873
},
{
"name": "transformer.h.18.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.18.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.18.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00831
},
{
"name": "transformer.h.18.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.0082
},
{
"name": "transformer.h.18.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.0082
},
{
"name": "transformer.h.19.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.19.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00624,
"zero_fraction": 0.00874
},
{
"name": "transformer.h.19.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0061,
"zero_fraction": 0.00867
},
{
"name": "transformer.h.19.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.19.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0061,
"zero_fraction": 0.00854
},
{
"name": "transformer.h.19.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00623,
"zero_fraction": 0.00911
},
{
"name": "transformer.h.19.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.19.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.19.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.19.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00595,
"zero_fraction": 0.00817
},
{
"name": "transformer.h.19.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00833
},
{
"name": "transformer.h.2.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.2.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00608,
"zero_fraction": 0.00826
},
{
"name": "transformer.h.2.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00825
},
{
"name": "transformer.h.2.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.2.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00607,
"zero_fraction": 0.00828
},
{
"name": "transformer.h.2.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.50744,
"zero_fraction": 0.45421
},
{
"name": "transformer.h.2.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.2.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.2.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.0083
},
{
"name": "transformer.h.2.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.00828
},
{
"name": "transformer.h.2.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.20.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.20.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00621,
"zero_fraction": 0.00875
},
{
"name": "transformer.h.20.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00843
},
{
"name": "transformer.h.20.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.20.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00611,
"zero_fraction": 0.00875
},
{
"name": "transformer.h.20.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00612,
"zero_fraction": 0.00929
},
{
"name": "transformer.h.20.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.20.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.20.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00828
},
{
"name": "transformer.h.20.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.0082
},
{
"name": "transformer.h.20.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00827
},
{
"name": "transformer.h.21.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.21.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00607,
"zero_fraction": 0.00872
},
{
"name": "transformer.h.21.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00606,
"zero_fraction": 0.00857
},
{
"name": "transformer.h.21.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.21.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00606,
"zero_fraction": 0.00855
},
{
"name": "transformer.h.21.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00609,
"zero_fraction": 0.00849
},
{
"name": "transformer.h.21.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.21.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.21.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00605,
"zero_fraction": 0.00832
},
{
"name": "transformer.h.21.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00819
},
{
"name": "transformer.h.21.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.22.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.22.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00631,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.22.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00807
},
{
"name": "transformer.h.22.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.22.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.00819
},
{
"name": "transformer.h.22.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00827
},
{
"name": "transformer.h.22.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.22.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.22.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0061,
"zero_fraction": 0.00829
},
{
"name": "transformer.h.22.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00826
},
{
"name": "transformer.h.22.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00828
},
{
"name": "transformer.h.23.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.23.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00608,
"zero_fraction": 0.00827
},
{
"name": "transformer.h.23.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00845
},
{
"name": "transformer.h.23.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.23.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00606,
"zero_fraction": 0.00828
},
{
"name": "transformer.h.23.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00856
},
{
"name": "transformer.h.23.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.23.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.23.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0063,
"zero_fraction": 0.00878
},
{
"name": "transformer.h.23.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.00839
},
{
"name": "transformer.h.23.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00815
},
{
"name": "transformer.h.3.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.3.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00607,
"zero_fraction": 0.00849
},
{
"name": "transformer.h.3.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00849
},
{
"name": "transformer.h.3.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.3.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.56119,
"zero_fraction": 0.44774
},
{
"name": "transformer.h.3.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00853
},
{
"name": "transformer.h.3.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.3.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.3.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00605,
"zero_fraction": 0.00833
},
{
"name": "transformer.h.3.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00822
},
{
"name": "transformer.h.3.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.006,
"zero_fraction": 0.0084
},
{
"name": "transformer.h.4.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.4.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00875
},
{
"name": "transformer.h.4.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00799
},
{
"name": "transformer.h.4.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.4.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.55556,
"zero_fraction": 0.44876
},
{
"name": "transformer.h.4.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00605,
"zero_fraction": 0.00948
},
{
"name": "transformer.h.4.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.4.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.4.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00602,
"zero_fraction": 0.00835
},
{
"name": "transformer.h.4.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00821
},
{
"name": "transformer.h.4.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.00828
},
{
"name": "transformer.h.5.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.5.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00612,
"zero_fraction": 0.00854
},
{
"name": "transformer.h.5.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00594,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.5.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.5.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.53754,
"zero_fraction": 0.44802
},
{
"name": "transformer.h.5.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00834
},
{
"name": "transformer.h.5.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.5.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.5.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00813
},
{
"name": "transformer.h.5.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00595,
"zero_fraction": 0.0081
},
{
"name": "transformer.h.5.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00826
},
{
"name": "transformer.h.6.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.6.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00835
},
{
"name": "transformer.h.6.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00825
},
{
"name": "transformer.h.6.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.6.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.54333,
"zero_fraction": 0.44813
},
{
"name": "transformer.h.6.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00863
},
{
"name": "transformer.h.6.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.6.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.6.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.00824
},
{
"name": "transformer.h.6.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00812
},
{
"name": "transformer.h.6.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00836
},
{
"name": "transformer.h.7.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.7.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.52404,
"zero_fraction": 0.44938
},
{
"name": "transformer.h.7.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.50169,
"zero_fraction": 0.44117
},
{
"name": "transformer.h.7.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.7.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.5452,
"zero_fraction": 0.44963
},
{
"name": "transformer.h.7.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "",
"relative_error": 0.49302,
"zero_fraction": 0.45467
},
{
"name": "transformer.h.7.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.7.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.7.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00822
},
{
"name": "transformer.h.7.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00595,
"zero_fraction": 0.0082
},
{
"name": "transformer.h.7.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00819
},
{
"name": "transformer.h.8.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.8.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0062,
"zero_fraction": 0.00875
},
{
"name": "transformer.h.8.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00833
},
{
"name": "transformer.h.8.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.8.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00606,
"zero_fraction": 0.00835
},
{
"name": "transformer.h.8.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00601,
"zero_fraction": 0.00876
},
{
"name": "transformer.h.8.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.8.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.8.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00829
},
{
"name": "transformer.h.8.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00595,
"zero_fraction": 0.00824
},
{
"name": "transformer.h.8.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00597,
"zero_fraction": 0.00826
},
{
"name": "transformer.h.9.attn.k_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.9.attn.k_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00613,
"zero_fraction": 0.00869
},
{
"name": "transformer.h.9.attn.o_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00596,
"zero_fraction": 0.00849
},
{
"name": "transformer.h.9.attn.q_norm.weight",
"shape": [
80
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.9.attn.q_proj.weight",
"shape": [
640,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.00824
},
{
"name": "transformer.h.9.attn.v_proj.weight",
"shape": [
320,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00603,
"zero_fraction": 0.00845
},
{
"name": "transformer.h.9.ln_1.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.9.ln_2.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.h.9.mlp.w_down.weight",
"shape": [
640,
1920
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00604,
"zero_fraction": 0.00824
},
{
"name": "transformer.h.9.mlp.w_gate.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00599,
"zero_fraction": 0.0082
},
{
"name": "transformer.h.9.mlp.w_up.weight",
"shape": [
1920,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.00598,
"zero_fraction": 0.00823
},
{
"name": "transformer.ln_f.weight",
"shape": [
640
],
"quantized": false,
"reason": "not a matrix",
"relative_error": 0.0,
"zero_fraction": 0.0
},
{
"name": "transformer.wte.weight",
"shape": [
32768,
640
],
"quantized": true,
"reason": "8-bit",
"relative_error": 0.0065,
"zero_fraction": 0.00893
}
]
}