Datasets:
w-ahmad
/
eval-test3

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
6.97k
[2026-06-11 13:12:58]
============================================================
[2026-06-11 13:12:58] GGUF MODEL EVALUATION
[2026-06-11 13:12:58] ============================================================
[2026-06-11 13:12:58] === CONFIG START ===
[2026-06-11 13:12:58] base_repo: unsloth/Qwen3.5-4B-GGUF
[2026-06-11 13:12:58] base_file: Qwen3.5-4B-BF16.gguf
[2026-06-11 13:12:58] ctx_size: 512
[2026-06-11 13:12:58] batch_size: 512
[2026-06-11 13:12:58] n_parallel: 1
[2026-06-11 13:12:58] seed: 1337
[2026-06-11 13:12:58] chunks: 50
[2026-06-11 13:12:58] dataset_repo: None
[2026-06-11 13:12:58] dataset_split: None
[2026-06-11 13:12:58] test_files: []
[2026-06-11 13:12:58] target_models: [["w-ahmad/test3", "MoQ-layers_0-3-IQ1_M/Qwen3.5_4B-MoQ-layers_0-3-IQ1_M.gguf"], ["w-ahmad/test3", "MoQ-layers_0-3-IQ1_S/Qwen3.5_4B-MoQ-layers_0-3-IQ1_S.gguf"], ["w-ahmad/test3", "MoQ-layers_0-3-IQ2_S/Qwen3.5_4B-MoQ-layers_0-3-IQ2_S.gguf"], ["w-ahmad/test3", "MoQ-layers_0-3-IQ2_XS/Q...
[2026-06-11 13:12:58] status: NEW
[2026-06-11 13:12:58] === CONFIG END ===
[2026-06-11 13:12:58] Eval dir : models/evaluation
[2026-06-11 13:12:58] Temp dir : /kaggle/working/
[2026-06-11 13:12:58] DAT dir : /tmp/llama_dat
[2026-06-11 13:12:58] Binary dir: /workspace/llama_bins/cuda-12.8
[2026-06-11 13:12:58] Binary : /workspace/llama_bins/cuda-12.8/llama-perplexity
[2026-06-11 13:12:58] Target models to evaluate: 83
[2026-06-11 13:12:58] Using default test file: /tmp/wiki_test.raw
[2026-06-11 13:12:58] Test files: 1
[2026-06-11 13:12:58]
📥 Downloading base model: unsloth/Qwen3.5-4B-GGUF / Qwen3.5-4B-BF16.gguf
[2026-06-11 13:13:25] ✅ Base model downloaded: /kaggle/working/Qwen3.5-4B-BF16.gguf
[2026-06-11 13:13:25]
============================================================
[2026-06-11 13:13:25] PHASE 1: Generating base logits for all files
[2026-06-11 13:13:25] ============================================================
[2026-06-11 13:13:25] [1/1] Base model on /tmp/wiki_test.raw → /tmp/llama_dat/base_logits_0.dat
[2026-06-11 13:13:25] Running: /workspace/llama_bins/cuda-12.8/llama-perplexity -m /kaggle/working/Qwen3.5-4B-BF16.gguf -f /tmp/wiki_test.raw -c 512 -b 512 -np 1 --kl-divergence-base /tmp/llama_dat/base_logits_0.dat --seed 1337 -ngl 99 --chunks 50
[2026-06-11 13:13:26] 0.00.286.022 I common_init_result: fitting params to device memory ...
[2026-06-11 13:13:26] 0.00.286.031 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
[2026-06-11 13:13:36] 0.10.566.612 W llama_context: n_ctx_seq (512) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
[2026-06-11 13:13:37] 0.10.617.541 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
[2026-06-11 13:14:31] 1.05.415.249 I
[2026-06-11 13:14:31] 1.05.415.442 I system_info: n_threads = 2 (n_threads_batch = 2) / 4 | CUDA : ARCHS = 750,800,860,890,900,1000,1200 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
[2026-06-11 13:14:31] 1.05.415.578 I perplexity: saving all logits to /tmp/llama_dat/base_logits_0.dat
[2026-06-11 13:14:31] 1.05.415.585 I perplexity: tokenizing the input ..
[2026-06-11 13:14:32] 1.06.250.582 I perplexity: tokenization took 834.984 ms
[2026-06-11 13:14:32] 1.06.251.068 I perplexity: calculating perplexity over 50 chunks, n_ctx=512, batch_size=512, n_seq=1
[2026-06-11 13:14:49] 1.23.550.020 I perplexity: 17.12 seconds per pass - ETA 14.25 minutes
[2026-06-11 13:16:38] [1]5.9954,[2]8.8643,[3]9.0343,[4]8.8617,[5]8.9319,[6]9.2448,[7]9.5683,[8]10.4290,[9]11.0487,[10]11.4813,[11]11.3769,[12]11.5143,[13]12.1970,[14]11.6112,[15]11.4999,[16]11.6183,[17]10.8196,[18]10.8113,[19]10.7109,[20]10.7439,[21]10.3575,[22]10.2251,[23]9.8303,[24]9.3419,[25]9.1660,[26]8.9160,[27]8....
[2026-06-11 13:16:39] 3.12.732.058 I Final estimate: PPL = 9.3905 +/- 0.22827
[2026-06-11 13:16:39]
[2026-06-11 13:16:40]
Phase 1 complete. Valid files: 1 / 1
[2026-06-11 13:16:41] 🗑️ Deleted /kaggle/working/Qwen3.5-4B-BF16.gguf to free disk space
[2026-06-11 13:16:41] ============================================================
[2026-06-11 13:16:41] PHASE 2: Running KL divergence for each target model
[2026-06-11 13:16:41] ============================================================
[2026-06-11 13:16:41]
============================================================
[2026-06-11 13:16:41] TARGET MODEL 1/83: w-ahmad/test3 / MoQ-layers_0-3-IQ1_M/Qwen3.5_4B-MoQ-layers_0-3-IQ1_M.gguf
[2026-06-11 13:16:41] ============================================================
[2026-06-11 13:17:25]
[1/1] MoQ-layers_0-3-IQ1_M/Qwen3.5_4B-MoQ-layers_0-3-IQ1_M.gguf vs /tmp/wiki_test.raw (base: /tmp/llama_dat/base_logits_0.dat)
[2026-06-11 13:17:25] Running: /workspace/llama_bins/cuda-12.8/llama-perplexity -m /kaggle/working/MoQ-layers_0-3-IQ1_M/Qwen3.5_4B-MoQ-layers_0-3-IQ1_M.gguf -f /tmp/wiki_test.raw -c 512 -b 512 -np 1 --kl-divergence-base /tmp/llama_dat/base_logits_0.dat --kl-divergence --seed 1337 -ngl 99 --chunks 50
[2026-06-11 13:17:26] 0.00.292.938 I common_init_result: fitting params to device memory ...
[2026-06-11 13:17:26] 0.00.292.974 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
[2026-06-11 13:17:36] 0.10.576.618 W llama_context: n_ctx_seq (512) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
[2026-06-11 13:17:36] 0.10.626.449 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
[2026-06-11 13:17:51] 0.25.871.733 I
[2026-06-11 13:17:51] 0.25.871.920 I system_info: n_threads = 2 (n_threads_batch = 2) / 4 | CUDA : ARCHS = 750,800,860,890,900,1000,1200 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
[2026-06-11 13:17:52] 0.25.963.375 I kl_divergence: computing over 50 chunks, n_ctx=512, batch_size=512, n_seq=1
[2026-06-11 13:17:54] 0.28.113.780 I kl_divergence: 2.15 seconds per pass - ETA 1.78 minutes
[2026-06-11 13:17:54]
[2026-06-11 13:17:54] chunk PPL ln(PPL(Q)/PPL(base)) KL Divergence Δp RMS Same top p
[2026-06-11 13:17:54] 1 8.3944 ± 1.5704 0.33655 ± 0.07684 0.30746 ± 0.04605 17.205 ± 1.782 % 80.392 ± 2.491 %
[2026-06-11 13:17:56] 2 11.2522 ± 1.5315 0.23875 ± 0.05297 0.33448 ± 0.02765 15.396 ± 1.178 % 80.392 ± 1.760 %
[2026-06-11 13:17:58] 3 11.5893 ± 1.2585 0.24920 ± 0.05060 0.44754 ± 0.03487 17.277 ± 1.056 % 78.693 ± 1.481 %
[2026-06-11 13:18:00] 4 11.3679 ± 1.0459 0.24915 ± 0.04551 0.41708 ± 0.03027 16.647 ± 0.887 % 78.333 ± 1.291 %
[2026-06-11 13:18:02] 5 10.8274 ± 0.8601 0.19253 ± 0.03742 0.36952 ± 0.02459 15.536 ± 0.774 % 78.667 ± 1.148 %
[2026-06-11 13:18:04] 6 11.3137 ± 0.8096 0.20202 ± 0.03219 0.34705 ± 0.02089 15.164 ± 0.707 % 78.301 ± 1.054 %
[2026-06-11 13:18:06] 7 11.6973 ± 0.7731 0.20095 ± 0.02885 0.33222 ± 0.01851 14.773 ± 0.660 % 78.824 ± 0.967 %
[2026-06-11 13:18:08] 8 12.6211 ± 0.7945 0.19082 ± 0.02695 0.34318 ± 0.01711 14.606 ± 0.615 % 77.794 ± 0.920 %
[2026-06-11 13:18:10] 9 13.3406 ± 0.7916 0.18854 ± 0.02525 0.33709 ± 0.01569 14.529 ± 0.577 % 77.647 ± 0.870 %
[2026-06-11 13:18:12] 10 13.7186 ± 0.7704 0.17806 ± 0.02352 0.33065 ± 0.01442 14.367 ± 0.548 % 77.529 ± 0.827 %
[2026-06-11 13:18:14] 11 13.4473 ± 0.7141 0.16722 ± 0.02191 0.32316 ± 0.01323 14.166 ± 0.518 % 77.825 ± 0.785 %
[2026-06-11 13:18:17] 12 13.6680 ± 0.6966 0.17149 ± 0.02099 0.32948 ± 0.01279 14.453 ± 0.500 % 77.353 ± 0.757 %
[2026-06-11 13:18:19] 13 14.2530 ± 0.7047 0.15580 ± 0.02083 0.33856 ± 0.01215 14.309 ± 0.471 % 76.772 ± 0.734 %
[2026-06-11 13:18:21] 14 13.4821 ± 0.6371 0.14985 ± 0.01965 0.33896 ± 0.01199 14.435 ± 0.449 % 76.947 ± 0.705 %
[2026-06-11 13:18:23] 15 13.1171 ± 0.5971 0.13201 ± 0.01899 0.34441 ± 0.01375 14.467 ± 0.446 % 77.307 ± 0.677 %
[2026-06-11 13:18:25] 16 13.2289 ± 0.5842 0.13022 ± 0.01852 0.34652 ± 0.01318 14.646 ± 0.435 % 77.132 ± 0.658 %
[2026-06-11 13:18:27] 17 12.2582 ± 0.5207 0.12522 ± 0.01765 0.33700 ± 0.01254 14.532 ± 0.421 % 77.670 ± 0.633 %
[2026-06-11 13:18:30] 18 12.3189 ± 0.5072 0.13090 ± 0.01732 0.33975 ± 0.01232 14.763 ± 0.416 % 77.560 ± 0.616 %
[2026-06-11 13:18:32] 19 12.2824 ± 0.4943 0.13724 ± 0.01703 0.34640 ± 0.01219 15.049 ± 0.413 % 77.523 ± 0.600 %
[2026-06-11 13:18:34] 20 12.1555 ± 0.4750 0.12376 ± 0.01647 0.34516 ± 0.01182 14.989 ± 0.405 % 77.647 ± 0.583 %
[2026-06-11 13:18:36] 21 11.7887 ± 0.4470 0.12974 ± 0.01586 0.34146 ± 0.01133 15.036 ± 0.392 % 77.759 ± 0.568 %
[2026-06-11 13:18:38] 22 11.5917 ± 0.4281 0.12574 ± 0.01534 0.33941 ± 0.01088 15.003 ± 0.378 % 77.825 ± 0.555 %
[2026-06-11 13:18:41] 23 11.1510 ± 0.4002 0.12634 ± 0.01482 0.33344 ± 0.01053 15.022 ± 0.371 % 77.954 ± 0.541 %
[2026-06-11 13:18:43] 24 10.6098 ± 0.3696 0.12754 ± 0.01429 0.32597 ± 0.01015 15.054 ± 0.363 % 78.382 ± 0.526 %
[2026-06-11 13:18:45] 25 10.4350 ± 0.3552 0.12992 ± 0.01396 0.32551 ± 0.00983 15.193 ± 0.351 % 78.416 ± 0.515 %
[2026-06-11 13:18:47] 26 10.1671 ± 0.3367 0.13156 ± 0.01361 0.32146 ± 0.00955 15.204 ± 0.344 % 78.673 ± 0.503 %
[2026-06-11 13:18:50] 27 10.0060 ± 0.3237 0.13453 ± 0.01326 0.32046 ± 0.00932 15.314 ± 0.336 % 78.693 ± 0.494 %