# Regenerate: python3 tools/int8_gemm/p16c6_same_op_c0_vs_c6.py --trace <--save-trace tensor>
# No GPU. Verified byte-identical to this file after the tool was rewritten.
PER-LINEAR: the SAME op at chunk 6 against ITSELF at chunk 0.
Row = (chunk, class, rel i8-vs-legB, rel i8-vs-fp32, rel legB-vs-fp32, M, K, N).
total scored int8 calls on this prompt: 3200
linears scored at BOTH c0 and c6: 400  (a positive count, not a filtered subset)
c6/c0 ratio over all 400: min=0.8623 p05=0.9508 p50=1.0139 p95=1.0697 max=1.1428

the layers the profile names, every linear in them:
class                                           c0 i8/f32  c6 i8/f32   c6/c0
model.layers.33.linear_attn.in_proj_qkv         1.278e-02  1.314e-02   1.028
model.layers.33.linear_attn.in_proj_z           1.409e-02  1.437e-02   1.019
model.layers.33.linear_attn.out_proj            1.712e-02  1.780e-02   1.040
model.layers.33.mlp.down_proj                   1.281e-02  1.268e-02   0.990
model.layers.33.mlp.gate_proj                   1.225e-02  1.246e-02   1.017
model.layers.33.mlp.up_proj                     1.106e-02  1.103e-02   0.997
model.layers.34.linear_attn.in_proj_qkv         1.288e-02  1.298e-02   1.008
model.layers.34.linear_attn.in_proj_z           1.357e-02  1.374e-02   1.012
model.layers.34.linear_attn.out_proj            1.516e-02  1.504e-02   0.992
model.layers.34.mlp.down_proj                   1.322e-02  1.336e-02   1.010
model.layers.34.mlp.gate_proj                   1.289e-02  1.304e-02   1.012
model.layers.34.mlp.up_proj                     1.131e-02  1.105e-02   0.977
model.layers.35.mlp.down_proj                   1.406e-02  1.345e-02   0.957
model.layers.35.mlp.gate_proj                   1.180e-02  1.247e-02   1.057
model.layers.35.mlp.up_proj                     1.153e-02  1.184e-02   1.027
model.layers.35.self_attn.k_proj                1.150e-02  1.160e-02   1.008
model.layers.35.self_attn.o_proj                1.924e-02  1.906e-02   0.990
model.layers.35.self_attn.q_proj                1.296e-02  1.369e-02   1.056
model.layers.35.self_attn.v_proj                1.094e-02  1.161e-02   1.061
model.layers.47.mlp.down_proj                   1.363e-02  1.356e-02   0.995
model.layers.47.mlp.gate_proj                   1.180e-02  1.215e-02   1.030
model.layers.47.mlp.up_proj                     1.158e-02  1.162e-02   1.003
model.layers.47.self_attn.k_proj                1.265e-02  1.250e-02   0.989
model.layers.47.self_attn.o_proj                1.986e-02  1.959e-02   0.986
model.layers.47.self_attn.q_proj                1.143e-02  1.245e-02   1.089
model.layers.47.self_attn.v_proj                1.039e-02  1.041e-02   1.002
model.layers.48.linear_attn.in_proj_qkv         1.296e-02  1.302e-02   1.005
model.layers.48.linear_attn.in_proj_z           1.317e-02  1.336e-02   1.015
model.layers.48.linear_attn.out_proj            1.932e-02  2.128e-02   1.101
model.layers.48.mlp.down_proj                   1.418e-02  1.434e-02   1.011
model.layers.48.mlp.gate_proj                   1.139e-02  1.168e-02   1.025
model.layers.48.mlp.up_proj                     1.128e-02  1.120e-02   0.993
model.layers.49.linear_attn.in_proj_qkv         1.255e-02  1.273e-02   1.014
model.layers.49.linear_attn.in_proj_z           1.293e-02  1.342e-02   1.038
model.layers.49.linear_attn.out_proj            1.794e-02  1.791e-02   0.998
model.layers.49.mlp.down_proj                   1.559e-02  1.660e-02   1.065
model.layers.49.mlp.gate_proj                   1.155e-02  1.177e-02   1.019
model.layers.49.mlp.up_proj                     1.162e-02  1.146e-02   0.986
model.layers.50.linear_attn.in_proj_qkv         1.227e-02  1.239e-02   1.010
model.layers.50.linear_attn.in_proj_z           1.212e-02  1.244e-02   1.027
model.layers.50.linear_attn.out_proj            1.881e-02  1.897e-02   1.008
model.layers.50.mlp.down_proj                   1.487e-02  1.452e-02   0.977
model.layers.50.mlp.gate_proj                   1.158e-02  1.175e-02   1.015
model.layers.50.mlp.up_proj                     1.141e-02  1.141e-02   1.000

the 12 linears with the LARGEST c6/c0 ratio ANYWHERE in the model:
  model.layers.62.linear_attn.in_proj_z           9.016e-03  1.030e-02   1.143
  model.layers.23.self_attn.q_proj                1.036e-02  1.172e-02   1.131
  model.layers.58.linear_attn.out_proj            1.895e-02  2.113e-02   1.115
  model.layers.48.linear_attn.out_proj            1.932e-02  2.128e-02   1.101
  model.layers.19.self_attn.q_proj                1.029e-02  1.133e-02   1.101
  model.layers.60.linear_attn.in_proj_z           9.029e-03  9.892e-03   1.096
  model.layers.4.linear_attn.out_proj             1.798e-02  1.969e-02   1.095
  model.layers.47.self_attn.q_proj                1.143e-02  1.245e-02   1.089
  model.layers.11.self_attn.q_proj                1.221e-02  1.324e-02   1.084
  model.layers.40.mlp.gate_proj                   1.158e-02  1.250e-02   1.079
  model.layers.62.linear_attn.in_proj_qkv         8.850e-03  9.526e-03   1.076
  model.layers.31.self_attn.q_proj                1.216e-02  1.307e-02   1.074
