Hugging Face's logo

Datasets:
junidude14
/
PBISC-data

Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
single-cell-rna-seq
bulk-rna-seq
diffusion
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
 Community
1
junidude14 commited on
Commit
173db0c
·
verified ·
1 Parent(s): e1f2c0c

Archive 4 checkpoints (1-4)

Browse files
Files changed (4) hide show
  1. paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/dataset/server_1/model8/plans/f01_trunk4x_complete_20261013/PLAN.md +78 -0
  2. paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/dataset/server_1/model8/plans/f01_trunk4x_complete_20261013/WORKFLOW_SEAL.json +105 -0
  3. paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/metadata/inference_locations.json +5 -0
  4. paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/metadata/request.json +91 -0
paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/dataset/server_1/model8/plans/f01_trunk4x_complete_20261013/PLAN.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 현재 논문 대표 PBISC와 동일한 6k panel로 trunk ×4 재학습
2
+
3
+ 2026-10-13 사용자 지시: gene context를 현재 ×1 checkpoint와 정확히 일치시켜 새로 pretraining하고, 논문 비교용 데이터를 준비한다. server_2 우선, 사용 중이면 server_1.
4
+
5
+ ## 고정할 것과 바꿀 것
6
+
7
+ 기준은 `f01_6k_1x_seed20260627/checkpoint_epoch_016.pt` (SHA `8c7cd0e83e2bc19fe03dfcbc084919a1e2f61be8cc8bc12f4caf543dfda91633`)다.
8
+
9
+ - **정확히 동일:** 6,000 gene ID와 순서, gene-selection manifest, 측정 mask 규칙, 33,141,380개 train cells, corpus split, conditioner, K recipe/top-up, batch256, worker4, prefetch32, seed20260627, bf16, lr0.0002, EMA0.999, gradient clipping5, 16 schedule parts, loss와 DDIM readout.
10
+ - **변경:** trunk `[1024,1024,512,256]` → `[4096,4096,2048,1024]`.
11
+ - **유지:** conditioner/context 폭512, time embedding128. Gene panel과 context-encoder 폭은 서로 다른 개념이다.
12
+ - Run ID/출력 경로와 CUDA allocator 상한8→24GiB는 운영상 변경이다. Host의98G/102G ML 제한은 높이지 않는다.
13
+ - 구 panel의 trunk×4 checkpoint를 warm start하지 않는다. 새 seed-controlled pretraining이다. 이전 Sepsis pilot은 별도 보존한다.
14
+ - 새 모델은 약240M parameters다. Downstream r16은 rank를 맞추며, adapter parameter 수까지 같다는 주장은 하지 않는다.
15
+
16
+ `CONTRACT.json`은 config 차이4개와 모든 주요 input/code hash를 기록한다. 5개 paired cohort의 저장된6k gene ID/order 일치와65개 기존 training-condition hash 일치를 확인했다.
17
+
18
+ ## 실행 위치와 저장소
19
+
20
+ 조사 시 두 서버 모두 사용 중이었다. 사용자 fallback에 따라 **server_1의 shared GPU lock에 대기 등록**했다. 기존 작업을 종료하지 않는다.
21
+
22
+ - Controller: `pbisc-f01-trunk4x-campaign-v2-20261013.service`
23
+ - GPU 실행: `ml-run` → 별도 `ml.slice` 작업. Guard active, swap0, OOMPolicy kill, OOMScoreAdjust+500. Pretraining job의 CPU memory high/max는64/72GiB, CUDA allocator는24GiB, trainer system-available floor는32GiB다.
24
+ - 먼저 worst-K **128-step profile**을 통과해야 본학습한다. OOM/guard/훈련 오류는 로그를 보존하고 중지하며 자동 재시작하지 않는다.
25
+ - 출력은 `payload/`에 연결된 NAS에 직접 쓴다. server_2의 기존 인증된 NAS mount를 경유하며, local/remote read-back SHA와 약3.56TiB 여유를 확인했다. 예상 payload 예산은1TiB다. 기존 원본을 삭제하지 않는다.
26
+ - 체크포인트는 매 part 보존한다. E16을 primary로 미리 고정하며, downstream 성능을 보고 checkpoint를 고르지 않는다. Dataset fingerprint, recipe/pool hash, emitted target order 및 총195,639,930 target occurrences가 기준×1과 맞아야 다음 단계로 간다.
27
+
28
+ ## 실험 행렬
29
+
30
+ | 단계 | 새 작업 | 비교/재사용 | Figure |
31
+ |---|---|---|---|
32
+ | 사전학습 | Profile128 + full16 parts 1회 | 현재×1의 동일 gene/context/stream contract | 1–2 |
33
+ | SLE | 기존과 같은5 sampling seeds;42 donors | ×1/scSemi 생성·annotation 재사용;1006/2279 평가축 구분 | 2 |
34
+ | Strict healthy | 5/10/20/40 total paired ×5 outer folds =20 LoRA | 추가 validation 없음, fixed3000, r16; 기존×1 동일 unit | 3 |
35
+ | Strict disease | 4 cohorts ×5 support draws =20 LoRA | 정확히5 samples/support, fixed3000; 기존×1/scSemi/deconv 재사용 | 4–5 |
36
+ | Historical CV | 5 cohorts ×5 folds =25 r16 LoRA | 원래 train/inner-val/test 유지 | 3–4 |
37
+ | Healthy modes | r4/r64/fullFT ×5 folds =15 | 원래 learning rate·early stopping·validation 접근 유지 | 3 |
38
+ | Legacy five-donor | 5 cohorts ×5 draws =25 route 중20개 새 adaptation | Sepsis5개는 strict와 동일하여 adapter와 generation 모두 재사용 | Supplementary/6 |
39
+ | Input controls | raw/refcal10개 + strict paired calibration40개 | 새 backbone 출력이므로 새 생성; 기존 calibration fit 재사용 | 3–4 |
40
+ | Bulk-only | Sepsis/Imperial 각5draw, 기존 row-parts/seed/readout 유지 | 해당 legacy five-donor adapter 연결; 기존×1 결과 재사용 | 6 |
41
+
42
+ **새 적응학습 총100회, paired route105개.** Sepsis5개 중복은 tensor-equivalent condition/seed까지 확인한 `GENERATION_REUSE_PROOF.json`으로 제외한다. 105번 새 학습하거나 같은 세포를 다시 생성하지 않는다. Source records와 전체 명령 입력은 `MATRIX.json`, `baseline_records/`에 고정했다.
43
+
44
+ ## 평가와 산출물
45
+
46
+ 1. 같은6k gene axis의 기존 scorer와 frozen1006-gene annotator를 사용한다. 이전 pilot의4258-gene 교집합 평가는 이번 primary에 사용하지 않는다.
47
+ 2. Precision/coverage, composition L1, donor expression r, cell-type logFC r, partial r, DEG, CAT/RBO와 추가 realism 지표를 계산한다. 질환 strict에는 donor-aware DEG를 별도로 계산한다.
48
+ 3. 같은 환자의 여러 visit/draw를 유지하는 paired donor bootstrap CI를 만든다. Support draw를 독립 환자 반복으로 간주하지 않는다.
49
+ 4. CV는5개 fold의 held-out 출력을 합친 뒤 pooled reference로 다시 평가한다. Fold P/C 평균을 pooled P/C로 바꾸어 표기하지 않는다.
50
+ 5. ×1 baseline의 test IDs, reference hash, contrast, gene exclusions가 다르면 비교를 보류하고 필요한 CPU rescore만 기록한다. Baseline 모델을 새로 학습하지 않는다.
51
+ 6. 생성 NPZ와 receipt, per-cell labels, per-donor metrics, donor-aware DEG, rare marker caches, bulk-only all6k cell-type pseudobulk/markers, pooled CV, 비교 CSV/PNG/PDF/SVG, checkpoint/입력/코드 hash를 보존한다.
52
+ 7. 기존 manuscript와154-page master는 자동 덮어쓰지 않는다. 새 결과 bundle을 검토한 뒤 figure별로 반영한다.
53
+
54
+ ## 완료와 연구적 판정의 구분
55
+
56
+ - Profile/16-part completion, input stream equality, 모든 예정 generation/score receipts가 있어야 새 모델 데이터가 준비됐다고 한다. 실패·missing·not-estimable은 그대로 남긴다.
57
+ - Figure5의 정확한 원 논문 population identity는 fine-cluster labels와 연령/감염 부위/steroid timing 검증이 필요하다. 새 marker cache만으로 재현 성공이라고 쓰지 않는다.
58
+ - Figure6는 bulk-only 데이터와 paired calibration 결과를 먼저 준비한다. Imperial HLA-II endpoint의 정확한 sample-time clinical contrast와 paired-reference 검증은 판정 gate로 남긴다. Generating cells와 published-finding reproduction은 다른 완료 상태다.
59
+ - 기존2k/3k/15k gene ablation은×1 family의 근거다. 이번6k×4 run으로 다른 gene-panel×4 실험까지 완료됐다고 바꾸지 않는다.
60
+ - K4000, training-seed 추가, 새로운 disease-state 탐색은 자동 확장하지 않는다. 특히 K1000 adapter를 K4000-adapted baseline으로 바꾸어 표기하지 않는다.
61
+
62
+ ## 진행 확인과 실패 시 처리
63
+
64
+ - `STATUS.json`: controller
65
+ - `PRETRAIN_STATUS.json`, `profile.log`, `train.log`: actual pretraining stage
66
+ - `payload/pretrain/.../RUN_STATUS.json`, `history.json`, `ALLCELL_DATA_CONTRACT.json`: 실제 학습 진도/coverage
67
+ - `DOWNSTREAM_STATUS.json`, `payload/runtime/receipts/`: 후속 작업
68
+ - `RESULTS_STATUS.json`, `PENDING_EVIDENCE.json`: 데이터 완료와 해석상 gate
69
+
70
+ 먼저 실패 로그와 guard/memory evidence를 확인한다. Resume는 원 run의 latest checkpoint와 고정 config를 사용하는 수동 복구로 제한한다. 기존 파일을 지우거나 새로운 seed로 실패를 덮지 않는다. 최종 아카이브 공간이 부족하면 NAS 원본을 보존한 채 archive pending으로 보고한다.
71
+
72
+ 실행 시간은128-step profile과 첫 실제 part의 측정값으로 추정한다. 본학습과100개 adaptation/생성까지 포함하므로 수일 단위가 될 수 있으며, 검증되지 않은 완료 시각은 약속하지 않는다.
73
+
74
+ 기존 corpus 내부 biological donor27명의 split 교차는 이번 비교에서 바꾸지 않는다. 같은 데이터 사용 경계를 유지한 width 비교이며, 외부 benchmark 누출을 새로 입증하거나 해소했다는 주장은 하지 않는다.
75
+
76
+ ## 2026-10-14 실행 변경
77
+
78
+ 사용자 지시로 pretraining 이후 두 서버에 고정 분담한다. 새 적응학습은 server_1/server_2 각각50회이며, 재사용5개는 원본 owner를 따른다. SLE/global/bulk-only 생성도 두 GPU에 배분하고 CPU 통계 평가는 server_1에서 순차 실행한다. 상세: `TWO_HOST_UPDATE.md`, `TWO_HOST_PLAN.json`. E16 직후 및 전체 계산 완료 시 핵심 checkpoint/결과를 HF에 해시 검증하여 보존한 뒤 독립 아카이브를 처리한다.
paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/dataset/server_1/model8/plans/f01_trunk4x_complete_20261013/WORKFLOW_SEAL.json ADDED
@@ -0,0 +1,105 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "files": {
3
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/pool_cv.py": "5e3778bce11399c4974caef85c38973ea33371615bbc5603f389edb90e718eaf",
4
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/build_matrix.py": "cf46d13015462bffe16decd6d70b7af9b3b54de826858019105882c85a382ee1",
5
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/score_sle.py": "071e6c0b24c82f47f418949aab888734481dc648973822c19a917b1a40edb6f9",
6
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/pretrain.py": "8941afeab49d78e10ec79d9f3aa85699644d7adbeb3e2d872e41a3096497ab5e",
7
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/finalize.py": "1bec014a0fd12ffcca4cf0e1ac8c81d02e4ebc4ba426b7a45b11fb2355540883",
8
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/score_paired.py": "ff290d84654d9f170b191ff9648b972b001c24fdcc04a8055958e06943e0e61c",
9
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/archive_completed.py": "2d15898f4c3cbc5136d541d5665277bba089ed09efd76f0922e1064b5f373c48",
10
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/bulk_compact.py": "75656af41adbb00e9574f631eb474f0f015711654894f40def9a2461a916daa3",
11
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/downstream.py": "32112a7ea74e06d63e812e521ad290ed8348e5585a32927c9ad9498769149645",
12
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/run_campaign.py": "211d0ca4721d74bcc902cad332204404760df9344431bf95eecb709d2eae103c",
13
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/prepare_legacy.py": "df9fe72593412df07564e828490a6661e6438f19599a512c3c853e4231445c20",
14
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/MATRIX.json": "7288cf39dd60540b6226cf81f34fc8a8c53aa6af390909115e4f221e0b35ea96",
15
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/GENERATION_REUSE_PROOF.json": "bd12c4e6c269155ebc89005d11499767e5e6e74438ea2fc8480c613ab09694d7",
16
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/ft_sdd.py": "9333b296acfb58e52ebe10e30c3334250bf2ce8c3dbfd446d65726f6b8da4abb",
17
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/scripts/sample_sdd_standalone.py": "95ace096cdc58a9ca92a18af649c60fe6ac0a55e97f977e7ab9bcb04efa968cb",
18
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/build_review_deck.py": "5bbfcf72211b8b29277da3ceba3ab16a91f7b64020232a1bcb9ca390d20e2f00",
19
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/figure5_panels.py": "f7b0b6887cd3d5219c9ceef825547920ed255e026685f6525b9d593afc7cca7a",
20
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/build_figure5.py": "5c0b6bc659e887f5d06265e8f49c7d4a798cc51b58d0fa1fa2d6fe324fe44466",
21
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/adopt_deconv_legacy.py": "cdb21114f08cf27e514960e8f30a63e17afb9b2c5d81bc100f2bad8b4b059b14",
22
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/campaign.py": "c0ed2ddf6ab13f4af1cd6bc785e52037a3e814b0af60965a0fef74d0d971adec",
23
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/prepare_strict_pairs.py": "0faf22ba69d8577344383a14569272207ae3dc507f9a3bcaa6eb6f274fe22d73",
24
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/finalize_outputs.py": "b03a4ddb579e416337f616119f435f5775ce80fdf1470d097d5db2f2fc11ca43",
25
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/assemble_manuscript.py": "65fe04e1a5ed3af24da625741382577d984c4d6023a9fcc66fabc369fbb4a7b5",
26
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/finalize_campaign.py": "12b0696b34163192c84392da2b29cb850fe9c7d5de3b14c3a2d14b9d618025da",
27
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/runtime.py": "67ea571aa077aa1d17131a9d5ade3eadc8c310620d8a358be81d3f69e065a7aa",
28
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/two_host_resume.py": "606b8de898959dc371904c2f1facde4974ebdf792412fcbe6b5ee1db9904cbd1",
29
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/export_scsemi_strict.py": "83c45ec6afc4d7d42dfb8317fb5c9645d1c9354daf39abc6de4dd96eef7fbd99",
30
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/rare_cache_builder.py": "c6f6c48a74853d803c224e97d113fbc76cf31e5773188facdcfcb558ff472cdb",
31
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/run_scsemi_strict.py": "5cfe000cf522b321ab1c35c894fcdefbf4c959f40cf1930132c5a4153351c433",
32
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/figure2_panels.py": "f3ed280558ab6034fdbf974f2b4262f0cde79216fed05b2b14aa09fae63db9c8",
33
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/export_healthy_visuals.py": "dc73ce329bc1609a84d4f80b103e61c7b6d8c4f0e939afee1963471a68ed7204",
34
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/build_figure1.py": "ab236b79c88d3784ec153cf9ff2b8c554810fcb5b5cdc3f8d06b1a34ae78d8a2",
35
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/archive_completed.py": "bfe368dd5b26cac3cbd0ba3d6381ad9c38249c88829ab7737b85a2eaa5607727",
36
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/deconv_strict.py": "2e1431e4bf5d44ca28b0d2f525a4fc225ab22899a88848080dda304e51372511",
37
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/rescore_pretrain_ablations.py": "843e3be4f236852e18903d9ba7f7ed4b4ab1c2680e3de9dccfab00d38f01dd90",
38
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/disease_signal.py": "23ae14b231e6417f2ec50ac4f1bffbc1a7fedcb975bfa8b3ad413f0ce6023acf",
39
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/prepare_controls.py": "e6c701bb35862565852c7ff6c647c9baf7ea103b6c33ff1d4decf2d262162bfb",
40
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/support_input_isolation.py": "762f5f7bbf908c478544b0c5c05a8fbf6c3bf63eff6a2d7af88105473c1c20a2",
41
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/build_results_figures.py": "959944ed0ebd38d859680709bbc0c185008efe36ea586a2712a6875c629426d1",
42
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/export_rare_visuals.py": "7dd66bd22318f99433cf3b387bbeee284140a8c38018b3b0e2f9d426b75e7f4b",
43
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/rare_metrics.py": "ab81ed9cd16b60a8efe868fdb0eb937ea077e261360a99bd4377b01b31ebd702",
44
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/verify_scsemi_adoption.py": "e8aab87ef624125d42489532d43a92985b2490792e5f45721f11d53049d94087",
45
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/support_manifest.py": "78185b8211f699cf3383897a79b369ba77644ddbbac1c1c948d27ee099d58c94",
46
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/figure5_details.py": "cd3aa31303dc0ae91a75f02aa1e4e1d271078e5df6f14b1d326cbdf04f6c1a3d",
47
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/donor_statistics.py": "78d99a9d44ad560c693460a3e57c23be48464892b4a941623336ed6c90dd174e",
48
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/score_strict.py": "bf6f264bfd377a6d9831ee42c553e58a6bcf44d5a926667eb75454b58dd34c42",
49
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/ablation_composition.py": "b4f6e882770cf7eb63cf059a1b060f08548b6e5595e20832ca1db6d996b8cd5c",
50
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/sample_one.py": "e1054d2e753361b33c44789eeae888234b700f4e3cafbba123948cd5c203dc04",
51
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/tools/import_scsemi_legacy.py": "da2bf6a0edd26a8d486a2bcc7fe9e554a91378ed22d28228dc3ae1045bbf4123",
52
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/build_archive_request.py": "3d89d044d7f1d60ea22c5b16097c089ae2d3be6c3405c2270e342fe17552c6b0",
53
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/prepare_fewshot.py": "944db7688b1ac97c4f8333f981226ed6922f3d55dd1861f141efcaca60c44d45",
54
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/contrast_extra_tables.py": "94f91a45048a09304e2340fdb4fb1d3e9f7958672a36013c8af29f5b21d493c1",
55
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/rescore_excluding_genes.py": "c69205bd9c23ccc42d1a9de6ebd5a6041ae7f137f85bba76c4f408a297dbc2fa",
56
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/aggregate_fewshot.py": "4dd3a622107a5dd4b07cefb46fee44c71522f41a06e046de35ce33eefa36bd21",
57
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/validate_immicro_published.py": "510bf22024d980b29ada6f3252443c5729db482cb79c06c243fc8a9463b6eb75",
58
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/build_full_cells.py": "055968f64c211d246848b2ba1de60282181b24039600589633c6eeceb4791b82",
59
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/check_rankcal.py": "d3e990006708b59628e0c18a8ac3a2273d0d6fec32e02f76bded7a54cbe3e19b",
60
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/build_tpm_zeroshot.py": "de19a1d7eccd8bb362eb934037d9ab65f6991e66edf8139940a6760ec96df50b",
61
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/pool_folds.py": "97ea51c014b2b5e2a3c458f390888924d573bb2dfb2e7fb196c373069b90412e",
62
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/aggregate_rankcal.py": "3cdfabcd4ba71e09dc85b3ce134a384c23475e192d3e6ea3dbc4c87e451b89df",
63
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/ft_sdd_aug.py": "487c9d13fff7e38ecfa4feb725077ec74a700cafd7b18dac5bc8abb35fb155f1",
64
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/score_ft.py": "ec5f40b4cea6d5062d0138f3b934abb6465e34cbfeac869c8a0906924c033aec",
65
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/build_sex_gene_list.py": "8637485d77bd6b85ae65423f59f7904dac0a3dfe3defcecaf13e49ab10e7724f",
66
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/prepare_cal_budget.py": "df2c07c6e129ddcf08ca45a1697e8b787f424ce05ed1b0253a0b4e96265d39a8",
67
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/build_immicro_cells.py": "caf366bf2ccaa7eada8b0df71a4d7faa8b2a84add546bc61d9da84c9388b0173",
68
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/contrast_extra.py": "b1ae7ded945177e61c1b2971c419c3adf5f9eb5e8868d427d1ba9d0f15fbf482",
69
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/aggregate.py": "5b5c57d90397540a34534bc4df5318b2b29a810f218e2007ff6ad0ea439646bc",
70
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/extra_metrics.py": "e76a6cd4c1c5ef7184e97755edcd0f48726b56ba251545a5b5a4b12bb766b0c2",
71
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/compare_scratch.py": "daa3d1f7b340980f7b9649a3716809659f54281175f8812067196141e4dcb5c8",
72
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/collect_extra_metrics.py": "6d403aeefaeec0537cbe58197beefb6ad17844b7058859c3c42c1cb1f7514a67",
73
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/ft_common.py": "4a6366b0f22208cd58714fc0261b8840c643b9d4569704d560cfd93534cbc747",
74
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/prepare_dataset.py": "36384262b23cf6260eff645812911c1a44f791ca8e4182765d91f2aabb3ccaef",
75
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/build_benchmark_queries.py": "681469d5d22cb97f38ceecc0fedec3081d71c9846ed44ec5447b86d27d818d4b",
76
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/build_reference.py": "329c13a6331511a58f9834dda273057daac7dfa6feae677c5fb62ac29dcc339b",
77
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/predict_scanvi.py": "482ee5dcf4d72b54bc36cbb6dcff62e5030bc7afc46cb5fb01eca73a82ee45a5",
78
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/build_queries_6k.py": "78da9c05d0b5bffbba6fe952725a7761d60b05ba350e84699979094df9dd8874",
79
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/verify_queries_2980.py": "44ec2d9aef8154bbb8c55b71944051328ea27237a7c3a6e54576b2008b8ad567",
80
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/build_truth_6k.py": "38f10525b3eea10db817b778fb3c3320dab38dd4b5e0418f306bcf00cefa4f13",
81
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/make_manifest.py": "5de1c4550bb99e87a52d69a0209b24d63e0379fc1c80542318525c9020b88bf0",
82
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/train_scanvi.py": "ba6c7b07cb24868c754e98b3508f77efc49d97ca59536af1b54da0193bd06e53",
83
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/scanvi_common.py": "b3e72157506611775f53793daf3c36ed397be348ed2e267fb789fdedc9b987d6",
84
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/evaluate_panel_6k.py": "2c7a4d62a93df942c002d868bf6f7f10aa120bdbbfc61a25e04a52bed2ab113a",
85
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/poisson_readout.py": "8bfa080942fbcdc89ae23b276524a1a83266fb8cbc56b9af279e50f82e6cde38",
86
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/report_scanvi2980.py": "7def021d3b7ac69915dfb2915ee5a39a9926a822e3d0279fc7a6c9c65b5bc60f",
87
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/panel_2980.py": "fa6cf72e523a38a3302181363e3f7cfd51c47f579d99757f3c94ffeba7156107",
88
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/build_reference_2980.py": "aa3c8b5ba3fb23223f2257d1ad70fa7e3a53ebf2da5b310942b88649f33e8ddf",
89
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/report_scanvi2980_extra.py": "c28c126f56b0cae775c2ca2fb25e27ee6277b0d817bb79b8194ba1f012e6b77f",
90
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/bulk2sc-paper/scripts/evaluate_generated_vs_truth.py": "48b12d794782b840838853e93e0fc55fab00532c4b3698f4588e926b985d2881",
91
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/bulkonly_lora5_20261006/tools/common.py": "2dd1be5960f9e3c3a4af42754ebe2609fcecb02e0ffe63ac6054700c30f17586",
92
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/dual_runtime.py": "e4366e5a2cc6c458ad1a96e950868638bec2455706bf39539ae00abc396716d8",
93
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/remote_gpu.py": "2b5876d7c53bfe26e600afa8b2b5bd3a253c4cdadfea4979a1462253f7a80898",
94
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/publish_core.py": "c38f4dabbf63364d46761a4b7cd45806dfa52f14f21cc12a86fd912c03a03e15",
95
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/TWO_HOST_PLAN.json": "1391b0debe7976f72763e8acf06c6fe43d730deeceb485aaca624a6147a23b87",
96
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/REMOTE_CODE_VERIFIED.json": "d0216d4016c94a077a0c675a9df03072862098f3a08b50684d9c496d969d5cdc",
97
+ "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/REMOTE_INPUTS_VERIFIED.json": "85d961fc98ee293b13fda8b4d2a0bfdc04ecb18dc53c8f03a9f6c1a4e9475764"
98
+ },
99
+ "no_silent_code_drift": true,
100
+ "authorized_transition": {
101
+ "reason": "User requested two-host downstream and HF backup 2026-10-14",
102
+ "previous_seal": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/revisions/20261014_before_two_host/WORKFLOW_SEAL.json",
103
+ "scientific_matrix_unchanged": true
104
+ }
105
+ }
paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/metadata/inference_locations.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "locations": [],
3
+ "project_id": "pbisc-diffusion",
4
+ "update_id": "20261013T223459Z-f01-two-host-dispatch"
5
+ }
paper-releases/pbisc-diffusion/updates/20261013T223459Z-f01-two-host-dispatch/diffusion/metadata/request.json ADDED
@@ -0,0 +1,91 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "assets": [
3
+ {
4
+ "host": "server_1",
5
+ "kind": "code",
6
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/downstream.py",
7
+ "state": "completed"
8
+ },
9
+ {
10
+ "host": "server_1",
11
+ "kind": "code",
12
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/dual_runtime.py",
13
+ "state": "completed"
14
+ },
15
+ {
16
+ "host": "server_1",
17
+ "kind": "code",
18
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/remote_gpu.py",
19
+ "state": "completed"
20
+ },
21
+ {
22
+ "host": "server_1",
23
+ "kind": "code",
24
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/publish_core.py",
25
+ "state": "completed"
26
+ },
27
+ {
28
+ "host": "server_1",
29
+ "kind": "code",
30
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/archive_completed.py",
31
+ "state": "completed"
32
+ },
33
+ {
34
+ "host": "server_1",
35
+ "kind": "code",
36
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/stage_remote.py",
37
+ "state": "completed"
38
+ },
39
+ {
40
+ "host": "server_1",
41
+ "kind": "code",
42
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/tools/test_two_host.py",
43
+ "state": "completed"
44
+ },
45
+ {
46
+ "host": "server_1",
47
+ "kind": "report",
48
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/TWO_HOST_PLAN.json",
49
+ "state": "completed"
50
+ },
51
+ {
52
+ "host": "server_1",
53
+ "kind": "report",
54
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/TWO_HOST_UPDATE.md",
55
+ "state": "completed"
56
+ },
57
+ {
58
+ "host": "server_1",
59
+ "kind": "report",
60
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/TWO_HOST_VERIFIED.json",
61
+ "state": "completed"
62
+ },
63
+ {
64
+ "host": "server_1",
65
+ "kind": "report",
66
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/REMOTE_INPUTS_VERIFIED.json",
67
+ "state": "completed"
68
+ },
69
+ {
70
+ "host": "server_1",
71
+ "kind": "report",
72
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/REMOTE_CODE_VERIFIED.json",
73
+ "state": "completed"
74
+ },
75
+ {
76
+ "host": "server_1",
77
+ "kind": "report",
78
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/WORKFLOW_SEAL.json",
79
+ "state": "completed"
80
+ },
81
+ {
82
+ "host": "server_1",
83
+ "kind": "report",
84
+ "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/f01_trunk4x_complete_20261013/PLAN.md",
85
+ "state": "completed"
86
+ }
87
+ ],
88
+ "inference_locations": [],
89
+ "project_id": "pbisc-diffusion",
90
+ "summary": "User-authorized post-pretraining two-host dispatch: 50 unique adaptation jobs per host, exact aliases preserved, guarded GPU execution and verified HF core backups before independent archive capacity gate. Training remains active and unmodified."
91
+ }