=== Quantization Report === Model: SpectreSystems/KAT-Coder-V2.5-Dev-MTP Date: 2026-08-08 09:16:13 Version: 0.14.3 CLI Arguments: --batch_size 8 --iters 1000 --nsamples 512 --seqlen 2048 --dataset opencode-instruct --output_dir ./models/KAT-Coder-V2.5-Dev-MTP-int4-AutoRound Memory Summary: Peak RAM: 68.53 GB Peak VRAM: 28.52 GB Sensitivity Analysis: ───────────────────────────────────────────────────────────────────────────────────────────────────── Layer Cosine Sim PSNR (dB) Iters uLoss Status ───────────────────────────────────────────────────────────────────────────────────────────────────── 🟢 model.language_model.layers.0 0.9994 62.7 780 11 → 0 PASS 🟢 model.language_model.layers.1 0.9994 63.1 996 3 → 1 PASS 🟢 model.language_model.layers.2 0.9994 66.9 993 7 → 1 PASS 🟢 model.language_model.layers.3 0.9990 62.2 846 9 → 2 PASS 🟢 model.language_model.layers.4 0.9989 61.6 812 6 → 3 PASS 🟢 model.language_model.layers.5 0.9987 60.8 972 8 → 4 PASS 🟢 model.language_model.layers.6 0.9987 71.9 874 9 → 4 PASS 🟢 model.language_model.layers.7 0.9979 67.7 979 15 → 7 PASS 🟢 model.language_model.layers.8 0.9977 66.3 912 13 → 8 PASS 🟢 model.language_model.layers.9 0.9976 65.4 814 15 → 9 PASS 🟢 model.language_model.layers.10 0.9980 80.2 810 23 → 11 PASS 🟢 model.language_model.layers.11 0.9973 79.1 985 29 → 14 PASS 🟢 model.language_model.layers.12 0.9970 78.9 899 24 → 15 PASS 🟢 model.language_model.layers.13 0.9965 77.9 812 27 → 18 PASS 🟢 model.language_model.layers.14 0.9969 77.3 976 35 → 22 PASS 🟢 model.language_model.layers.15 0.9948 75.6 674 51 → 30 PASS 🟢 model.language_model.layers.16 0.9945 74.8 886 57 → 36 PASS 🟢 model.language_model.layers.17 0.9935 74.3 467 61 → 39 PASS 🟢 model.language_model.layers.18 0.9932 72.9 839 81 → 55 PASS 🟢 model.language_model.layers.19 0.9909 71.4 918 114 → 77 PASS 🟢 model.language_model.layers.20 0.9909 71.3 649 106 → 81 PASS 🟢 model.language_model.layers.21 0.9906 71.0 824 116 → 80 PASS 🟠 model.language_model.layers.22 0.9895 69.4 592 168 → 117 WARN 🟠 model.language_model.layers.23 0.9868 68.7 394 192 → 141 WARN 🟠 model.language_model.layers.24 0.9862 68.7 694 173 → 140 WARN 🟠 model.language_model.layers.25 0.9856 68.1 924 230 → 164 WARN 🟠 model.language_model.layers.26 0.9872 67.9 576 223 → 170 WARN 🟠 model.language_model.layers.27 0.9850 67.1 586 266 → 211 WARN 🟠 model.language_model.layers.28 0.9864 66.8 786 306 → 225 WARN 🟠 model.language_model.layers.29 0.9865 66.9 828 307 → 217 WARN 🟠 model.language_model.layers.30 0.9873 66.0 587 353 → 267 WARN 🟠 model.language_model.layers.31 0.9853 48.2 803 447 → 329 WARN 🟠 model.language_model.layers.32 0.9851 46.9 783 469 → 354 WARN 🟠 model.language_model.layers.33 0.9869 47.7 231 653 → 442 WARN 🟠 model.language_model.layers.34 0.9869 58.4 806 864 → 558 WARN 🟠 model.language_model.layers.35 0.9870 48.3 916 1070 → 737 WARN 🟠 model.language_model.layers.36 0.9867 47.2 913 1211 → 870 WARN 🟠 model.language_model.layers.37 0.9886 48.1 770 1607 → 1098 WARN 🟠 model.language_model.layers.38 0.9893 55.8 491 2262 → 1553 WARN 🟢 model.language_model.layers.39 0.9918 54.0 814 4543 → 2399 PASS Summary: Total blocks: 40 Passed (🟢): 23 Warning (🟠): 17 Thresholds: Cosine Similarity < 0.99, PSNR < 45.0 dB