Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261006T033847Z-bulkonly-lora5-k4000/diffusion/dataset/server_1/model8/plans/bulkonly_lora5_k4000_20261006/logs/run_gpu_spark-fa6b_20261006T072546.log

@@ -0,0 +1,26 @@
+2026-10-06T07:25:46 runner start on spark-fa6b: ye2026_sepsis imperial_eskd
+available_gib=107.67 threshold_gib=18.00 ml_populated=yes cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T07:25:50 --- ye2026_sepsis draw0: ml-run gpu_draw.sh (ok: 63C; waits in the GPU lock while another job holds it)
+2026-10-06T08:02:14 ye2026_sepsis draw0: 7 sampling parts done; 2026-10-06T08:02:14 === ye2026_sepsis draw0 complete
+available_gib=110.86 threshold_gib=18.00 ml_populated=no cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T08:02:18 --- ye2026_sepsis draw1: ml-run gpu_draw.sh (ok: 57C; waits in the GPU lock while another job holds it)
+2026-10-06T08:31:18 ye2026_sepsis draw1: 7 sampling parts done; 2026-10-06T08:31:18 === ye2026_sepsis draw1 complete
+available_gib=106.58 threshold_gib=18.00 ml_populated=no cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T08:31:22 --- ye2026_sepsis draw2: ml-run gpu_draw.sh (cooled to 63C (<= 65C resume mark); releasing; waits in the GPU lock while another job holds it)
+2026-10-06T08:59:53 ye2026_sepsis draw2: 7 sampling parts done; 2026-10-06T08:59:53 === ye2026_sepsis draw2 complete
+available_gib=107.37 threshold_gib=18.00 ml_populated=no cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T08:59:57 --- ye2026_sepsis draw3: ml-run gpu_draw.sh (ok: 56C; waits in the GPU lock while another job holds it)
+2026-10-06T09:27:47 ye2026_sepsis draw3: 7 sampling parts done; 2026-10-06T09:27:47 === ye2026_sepsis draw3 complete
+available_gib=107.62 threshold_gib=18.00 ml_populated=no cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T09:27:51 --- ye2026_sepsis draw4: ml-run gpu_draw.sh (ok: 55C; waits in the GPU lock while another job holds it)
+2026-10-06T09:55:49 ye2026_sepsis draw4: 7 sampling parts done; 2026-10-06T09:55:49 === ye2026_sepsis draw4 complete
+available_gib=107.59 threshold_gib=18.00 ml_populated=no cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T09:55:53 --- imperial_eskd draw0: ml-run gpu_draw.sh (ok: 55C; waits in the GPU lock while another job holds it)
+2026-10-06T10:34:53 imperial_eskd draw0: 10 sampling parts done; 2026-10-06T10:34:53 === imperial_eskd draw0 complete
+available_gib=107.68 threshold_gib=18.00 ml_populated=no cgroup=/user.slice/user-1000.slice/user@1000.service/ml.slice
+2026-10-06T10:34:57 --- imperial_eskd draw1: ml-run gpu_draw.sh (ok: 56C; waits in the GPU lock while another job holds it)
+2026-10-06T11:13:34 imperial_eskd draw1: 10 sampling parts done; 2026-10-06T11:13:34 === imperial_eskd draw1 complete
+2026-10-06T11:13:34 imperial_eskd draw2: done already
+2026-10-06T11:13:34 imperial_eskd draw3: done already
+2026-10-06T11:13:34 imperial_eskd draw4: done already
+2026-10-06T11:13:34 all draws done

paper-releases/pbisc-diffusion/updates/20261006T033847Z-bulkonly-lora5-k4000/diffusion/dataset/server_1/model8/plans/bulkonly_lora5_k4000_20261006/logs/score_ye2026_sepsis_d0.log

@@ -0,0 +1,16 @@
+07:47:59 ye2026_sepsis: 77 scored donors, 2 arm(s); fitting the observed reference
+07:48:03 reference: R1 11550 cells, PCA var 0.336, R2 control P/C 0.9815/0.9997; contrast on
+07:48:03 observed top-50 DEG lists: T 50 (sig 1558, 57+20 donors, 281037 cells), B 50 (sig 1022, 57+20 donors, 94421 cells), Mono 50 (sig 2259, 57+20 donors, 106918 cells), NK 50 (sig 1095, 57+20 donors, 42712 cells), DC 50 (sig 1229, 54+20 donors, 3096 cells)
+07:48:03 lora5: loading generated_counts (115500 cells in the NPZ, 115500 scored)
+07:48:14 lora5: Poisson read-out nnz 33618971 (whole NPZ), median library 1873; annotating
+07:48:23 lora5: L1 0.4723 (null 0.5124478114694226) P/C 0.9268/0.7059 pb r 0.8780 top1 0.052 mono 11.2% (19.4 s)
+07:48:23 lora5: top-50 DEG 7/250 (T 1, B 5, Mono 1, NK 0, DC 0); broadcast 23 (0.21 s)
+07:48:23 lora5_k4000: loading generated_counts (115500 cells in the NPZ, 115500 scored)
+07:48:33 lora5_k4000: Poisson read-out nnz 33203310 (whole NPZ), median library 1776; annotating
+07:48:41 lora5_k4000: L1 0.4786 (null 0.5124478114694226) P/C 0.9179/0.6958 pb r 0.8793 top1 0.039 mono 9.6% (18.7 s)
+07:48:41 lora5_k4000: top-50 DEG 6/250 (T 1, B 2, Mono 3, NK 0, DC 0); broadcast 23 (0.21 s)
+             comp_l1  comp_l1_obs_loo_null  precision  coverage  precision_dm  coverage_dm  pb_r_median  pb_top1  mono_pct_gen  mono_pct_obs  median_umi_gen  median_umi_obs  contrast_mean_cp10k_r_gen  contrast_pseudobulk_log1p_cpm_r_gen  contrast_mean_cp10k_r_broadcast  contrast_pseudobulk_log1p_cpm_r_broadcast  deg_top50_T  deg_top50_B  deg_top50_Mono  deg_top50_NK  deg_top50_DC  deg_top50_sum  deg_top50_max  deg_top50_sum_4type  deg_top50_sum_broadcast  seconds
+lora5         0.4723                0.5124     0.9268    0.7059        0.9396       0.8258       0.8780   0.0519       11.2407       21.1554          1873.0          1966.0                     0.4360                               0.4203                            0.336                                     0.2989          1.0          5.0             1.0           0.0           0.0            7.0          250.0                  7.0                     23.0     19.4
+lora5_k4000   0.4786                0.5124     0.9179    0.6958        0.9360       0.8144       0.8793   0.0390        9.6000       21.1554          1776.0          1966.0                     0.4265                               0.4081                            0.336                                     0.2989          1.0          2.0             3.0           0.0           0.0            6.0          250.0                  6.0                     23.0     18.7
+07:48:41 done in 48.3 s -> /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/bulkonly_lora5_k4000_20261006/scores/ye2026_sepsis/draw0
+{"lora5": {"comp_l1": 0.4723, "coverage": 0.7059, "pb_r_median": 0.878, "contrast_mean_cp10k_r_gen": 0.436, "deg_top50_sum": 7}, "lora5_k4000": {"comp_l1": 0.4786, "coverage": 0.6958, "pb_r_median": 0.8793, "contrast_mean_cp10k_r_gen": 0.4265, "deg_top50_sum": 6}}

paper-releases/pbisc-diffusion/updates/20261006T033847Z-bulkonly-lora5-k4000/diffusion/dataset/server_1/model8/plans/bulkonly_lora5_k4000_20261006/logs/share_r2_20261006T122449.log

@@ -0,0 +1,11 @@
+2026-10-06T12:24:49 staged analysis: 1.8M
+2026-10-06T12:24:51 sha256 of 26 objects -> SHA256SUMS
+2026-10-06T12:24:51 uploading share/
+2026/10/06 12:25:53 NOTICE: S3 bucket pbisc-data-private path shares/bulkonly_lora5_k4000_20261006: 0 differences found
+2026/10/06 12:25:53 NOTICE: S3 bucket pbisc-data-private path shares/bulkonly_lora5_k4000_20261006: 15 matching files
+2026-10-06T12:25:53 uploading analysis/
+2026/10/06 12:25:55 NOTICE: S3 bucket pbisc-data-private path shares/bulkonly_lora5_k4000_20261006/analysis: 0 differences found
+2026/10/06 12:25:55 NOTICE: S3 bucket pbisc-data-private path shares/bulkonly_lora5_k4000_20261006/analysis: 12 matching files
+2026-10-06T12:25:56 all uploads checked (sizes)
+2026-10-06T12:25:56 making presigned links (expire 168h)
+2026-10-06T12:26:11 index uploaded; HTTP 200; 27 files linked; links valid until 2026-10-13 12:25 KST

paper-releases/pbisc-diffusion/updates/20261006T033847Z-bulkonly-lora5-k4000/diffusion/dataset/server_1/model8/plans/bulkonly_lora5_k4000_20261006/logs/ye2026_sepsis-aggregate.log

@@ -0,0 +1,26 @@
+# ye2026_sepsis: LoRA-5 at K = 4,000 vs K = 1,000 on the paired benchmark (5 draws)
+
+Same five few-pair draws (same 5 training donors each), same test donors, same scorer (paired_finetune_intradataset_20260927/tools/score_ft.py), 1,500 cells per test sample. Mean ± SD over draws.
+
+| metric | LoRA-5, K = 1,000 (benchmark) | LoRA-5, K = 4,000 | K4000 − K1000 (draws better / n) |
+|---|---:|---:|---|
+| comp_l1 | 0.466 ± 0.044 | 0.445 ± 0.051 | -0.022 (4 / 5) |
+| precision | 0.906 ± 0.020 | 0.910 ± 0.019 | +0.004 (2 / 5) |
+| coverage | 0.694 ± 0.042 | 0.698 ± 0.035 | +0.004 (3 / 5) |
+| coverage_dm | 0.834 ± 0.023 | 0.836 ± 0.023 | +0.001 (2 / 5) |
+| pb_r_median | 0.883 ± 0.003 | 0.884 ± 0.003 | +0.001 (3 / 5) |
+| mono_pct_gen | 17.839 ± 9.673 | 18.706 ± 10.052 | +0.867 (2 / 5) |
+| mono_pct_obs | 20.689 ± 0.444 | 20.689 ± 0.444 | +0.000 |
+| median_umi_gen | 1728.600 ± 189.579 | 1781.200 ± 169.896 | +52.600 (2 / 5) |
+| median_umi_obs | 1971.000 ± 18.841 | 1971.000 ± 18.841 | +0.000 |
+| deg_top50_sum | 16.400 ± 6.731 | 17.800 ± 9.445 | +1.400 (2 / 5) |
+| deg_top50_sum_broadcast | 20.000 ± 1.871 | 20.000 ± 1.871 | +0.000 |
+| logfc_r_gen | 0.350 ± 0.162 | 0.327 ± 0.128 | -0.023 (2 / 5) |
+| logfc_r_broadcast | 0.320 ± 0.024 | 0.320 ± 0.024 | +0.000 |
+| logfc_partial_r | 0.305 ± 0.189 | 0.276 ± 0.160 | -0.029 (2 / 5) |
+
+Per donor-draw composition L1, K = 4,000 − K = 1,000: -0.0215; K = 4,000 is better in 57% of 385 donor-draws (Wilcoxon p 0.0054).
+
+Lower is better for comp_l1; closer to *_obs is better for mono_pct_gen and median_umi_gen; higher is better for precision, coverage, pseudobulk r, the logFC r values and the top-50 DEG overlap. *_obs and *broadcast rows are the same for both arms (observed truth / bulk baseline).
+
+Scorer check: this plan's K = 1,000 arm vs the benchmark's stored draw scores, max |difference| per draw: draw0 0 (25 metrics), draw1 0 (25 metrics), draw2 0 (25 metrics), draw3 0 (25 metrics), draw4 0 (25 metrics).