| === Quantization Report === |
| Model: Qwen/Qwen3.6-27B |
| Date: 2026-06-21 16:31:05 |
| Version: 0.14.2 |
|
|
| CLI Arguments: |
| --batch_size 8 |
| --iters 1000 |
| --nsamples 512 |
| --seqlen 2048 |
| --dataset opencode-instruct |
| --output_dir ./models/Qwen3.6-27B-int4-AutoRound |
|
|
| Memory Summary: |
| Peak RAM: 36.13 GB |
| Peak VRAM: 43.13 GB |
|
|
| Sensitivity Analysis: |
| ───────────────────────────────────────────────────────────────────────────────────────────────────── |
| Layer Cosine Sim PSNR (dB) Iters uLoss Status |
| ───────────────────────────────────────────────────────────────────────────────────────────────────── |
| 🟢 model.language_model.layers.0 1.0000 81.8 947 52 → 3 PASS |
| 🟢 model.language_model.layers.1 1.0000 84.0 947 24 → 4 PASS |
| 🟢 model.language_model.layers.2 0.9999 83.6 844 26 → 8 PASS |
| 🟢 model.language_model.layers.3 0.9999 83.4 981 61 → 11 PASS |
| 🟢 model.language_model.layers.4 0.9999 82.9 761 33 → 13 PASS |
| 🟢 model.language_model.layers.5 0.9999 82.3 883 42 → 16 PASS |
| 🟢 model.language_model.layers.6 0.9999 82.5 834 84 → 23 PASS |
| 🟢 model.language_model.layers.7 0.9999 81.1 907 114 → 31 PASS |
| 🟢 model.language_model.layers.8 0.9999 80.7 550 80 → 33 PASS |
| 🟢 model.language_model.layers.9 0.9999 81.2 789 91 → 43 PASS |
| 🟢 model.language_model.layers.10 0.9999 82.8 823 115 → 53 PASS |
| 🟢 model.language_model.layers.11 0.9999 81.4 872 196 → 68 PASS |
| 🟢 model.language_model.layers.12 0.9999 80.5 946 190 → 80 PASS |
| 🟢 model.language_model.layers.13 0.9999 80.2 641 204 → 99 PASS |
| 🟢 model.language_model.layers.14 0.9998 81.5 965 268 → 122 PASS |
| 🟢 model.language_model.layers.15 0.9998 80.2 796 319 → 150 PASS |
| 🟢 model.language_model.layers.16 0.9998 79.3 967 343 → 165 PASS |
| 🟢 model.language_model.layers.17 0.9998 78.3 467 368 → 183 PASS |
| 🟢 model.language_model.layers.18 0.9997 79.6 839 546 → 252 PASS |
| 🟢 model.language_model.layers.19 0.9996 76.9 918 791 → 341 PASS |
| 🟢 model.language_model.layers.20 0.9995 74.9 951 824 → 408 PASS |
| 🟢 model.language_model.layers.21 0.9995 78.7 935 956 → 452 PASS |
| 🟢 model.language_model.layers.22 0.9995 78.2 682 1126 → 527 PASS |
| 🟢 model.language_model.layers.23 0.9994 77.7 792 1133 → 624 PASS |
| 🟢 model.language_model.layers.24 0.9993 77.4 906 1292 → 641 PASS |
| 🟢 model.language_model.layers.25 0.9993 77.1 415 1484 → 717 PASS |
| 🟢 model.language_model.layers.26 0.9993 76.4 342 1831 → 890 PASS |
| 🟢 model.language_model.layers.27 0.9991 75.9 998 2366 → 1028 PASS |
| 🟢 model.language_model.layers.28 0.9991 75.5 881 2113 → 1142 PASS |
| 🟢 model.language_model.layers.29 0.9990 75.1 869 2507 → 1180 PASS |
| 🟢 model.language_model.layers.30 0.9990 74.4 536 2671 → 1448 PASS |
| 🟢 model.language_model.layers.31 0.9988 73.6 992 3648 → 1718 PASS |
| 🟢 model.language_model.layers.32 0.9986 73.2 783 3364 → 1860 PASS |
| 🟢 model.language_model.layers.33 0.9986 73.0 958 3695 → 2074 PASS |
| 🟢 model.language_model.layers.34 0.9985 72.4 524 4149 → 2225 PASS |
| 🟢 model.language_model.layers.35 0.9981 71.2 814 5859 → 3104 PASS |
| 🟢 model.language_model.layers.36 0.9980 71.0 913 5088 → 3200 PASS |
| 🟢 model.language_model.layers.37 0.9979 70.5 969 6147 → 3629 PASS |
| 🟢 model.language_model.layers.38 0.9980 70.3 886 5801 → 3836 PASS |
| 🟢 model.language_model.layers.39 0.9977 69.6 961 6741 → 4432 PASS |
| 🟢 model.language_model.layers.40 0.9976 69.4 447 6766 → 4467 PASS |
| 🟢 model.language_model.layers.41 0.9976 69.2 967 7301 → 4725 PASS |
| 🟢 model.language_model.layers.42 0.9972 68.0 642 9277 → 5859 PASS |
| 🟢 model.language_model.layers.43 0.9968 67.1 759 13153 → 7260 PASS |
| 🟢 model.language_model.layers.44 0.9965 66.6 878 13606 → 7406 PASS |
| 🟢 model.language_model.layers.45 0.9965 66.2 963 12930 → 8683 PASS |
| 🟢 model.language_model.layers.46 0.9963 65.4 795 15271 → 9870 PASS |
| 🟢 model.language_model.layers.47 0.9960 63.7 927 17557 → 11729 PASS |
| 🟢 model.language_model.layers.48 0.9956 61.8 977 19889 → 12692 PASS |
| 🟢 model.language_model.layers.49 0.9954 61.0 459 23092 → 15256 PASS |
| 🟢 model.language_model.layers.50 0.9943 60.2 513 28450 → 20103 PASS |
| 🟢 model.language_model.layers.51 0.9934 54.9 836 57060 → 30481 PASS |
| 🟢 model.language_model.layers.52 0.9931 54.7 583 55960 → 33850 PASS |
| 🟢 model.language_model.layers.53 0.9926 54.5 225 75904 → 44263 PASS |
| 🟢 model.language_model.layers.54 0.9901 66.1 520 98444 → 71893 PASS |
| 🟢 model.language_model.layers.55 0.9915 63.5 567 101619 → 73226 PASS |
| 🟢 model.language_model.layers.56 0.9916 62.3 565 125887 → 83444 PASS |
| 🟢 model.language_model.layers.57 0.9916 62.0 359 110249 → 92079 PASS |
| 🟠 model.language_model.layers.58 0.9883 66.5 575 207342 → 126733 WARN |
| 🟢 model.language_model.layers.59 0.9923 56.3 869 184128 → 122327 PASS |
| 🟢 model.language_model.layers.60 0.9926 54.5 750 203348 → 142544 PASS |
| 🟢 model.language_model.layers.61 0.9928 54.6 663 222551 → 165940 PASS |
| 🟢 model.language_model.layers.62 0.9917 53.4 536 303288 → 210218 PASS |
| 🟢 model.language_model.layers.63 0.9925 58.5 544 486854 → 324412 PASS |
|
|
| Summary: |
| Total blocks: 64 |
| Passed (🟢): 63 |
| Warning (🟠): 1 |
|
|
| Thresholds: Cosine Similarity < 0.99, PSNR < 45.0 dB |
|
|
