Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261015T072909Z-disease-width-readiness/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/recoveries/disease_width_comparison_20261015/CHILI_TRANSFER_VERIFIED.json

@@ -0,0 +1,10 @@
+{
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/generated/ji2025_chili/disease/draw0/lora.npz": "fba94878b514efea1019e36ad51e83fb58a81305e8b692137775563d44548b09",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/generated/ji2025_chili/disease/draw0/lora.log": "92ec9b798db9f855576272e391ecb63c0a55e53bbf394a478383c9feae6daff2",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/runs/ji2025_chili/disease/draw0/lora/RUN.json": "9a34723dd42eb52a8cee11812d60017762a3c688989435325bba89c649f4d242",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/runs/ji2025_chili/disease/draw0/lora/checkpoint_ft.pt": "53488bb1133c0b5969d6094112798d7666e055f1dea3f630c8db1064d892a08b",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/generated/ji2025_chili/disease/draw1/lora.npz": "88f760546ae596ec9207d30d463b5912734e877356186833a4752c8925a6bc50",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/generated/ji2025_chili/disease/draw1/lora.log": "3c632e6ffb9b94e86b465c0306826aa8f8e2c14b4d66ce391134189b8fb65ef9",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/runs/ji2025_chili/disease/draw1/lora/RUN.json": "bf59fadc592f4bf0f6d67ae1b53307f3f88d25369b0e2ad417fe153d752d6f8d",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/runs/ji2025_chili/disease/draw1/lora/checkpoint_ft.pt": "a9a14ff9361e6cc530320810b94100e7a38353374fa54aa3fab7552c8cf4945f"
+}

paper-releases/pbisc-diffusion/updates/20261015T072909Z-disease-width-readiness/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/recoveries/disease_width_comparison_20261015/PLAN.md

@@ -0,0 +1,23 @@
+# Sepsis / Imperial / CHILI: matched ×1 versus ×4 LoRA
+
+2026-10-15. No new GPU experiment. Existing×1 outputs and existing×4 workflow are used unchanged.
+
+## Current readiness
+- Sepsis: x1 five generations and five prior summaries reused. x4 not yet generated.
+- Imperial: x1 five generations ready; four prior summaries reused, draw4 core metrics computed on CPU after provenance verification. x4 not yet generated.
+- CHILI: x1 draw0/1 generations imported from server_2 with8 required files SHA256 verified; both core summaries computed on CPU. Draw2–4 x1 work remains behind the x4 priority gate. x4 not yet generated.
+- No direct paired x1/x4 result exists for these three cohorts at the recorded snapshot. Existing Giroux figures must not be substituted.
+
+## Automatic comparison
+`pbisc-disease-width-cpu-20261015.service` checks readiness every120s. On each available draw, it checks support/train/test/seed/rank/alpha/learning rate/warmup/batch/EMA/fixed steps, generated gene order and sample IDs, condition hash, reference/annotation/contrast/gene exclusion/DEG definition before emitting MATCHED.json and COMPARISON.csv. It never launches training or sampling. Only missing x1 core score tables are computed, on CPU with2threads and16GiB process limit. Core comparisons need not wait for additional realism statistics.
+
+`pbisc-disease-width-archive-watch-20261015.service` uploads completed15-draw tables/provenance through the versioned paper archive/HF workflow. On a failure, preserve evidence and inspect; no compute restart loop.
+
+## Estimated availability, not deadlines
+Current x4 execution order is remaining healthy -> Imperial -> CHILI -> Sepsis. Healthy scoring has taken20–22min/unit; existing Imperial full scoring0.9–2.2h/unit and Sepsis0.5h/unit, with different arm counts/hosts, so estimates are approximate.
+- Imperial first matched draw: Oct15night–Oct16early morning; five draws: Oct16morning–afternoon.
+- CHILI first two matched draws: Oct16morning–afternoon. All five require remaining x1 runs after x4 core completion; currently Oct18or later is the conservative expectation, not a guarantee.
+- Sepsis first matched draw: Oct16afternoon–Oct17early morning; all five approximately Oct16night–Oct17, depending on CPU queue.
+No queue reorder or new GPU work was introduced to obtain these estimates.
+
+Descriptive support-draw means only; overlapping draws are not independent patient replicates. References to readiness are snapshots, not live completion claims.

paper-releases/pbisc-diffusion/updates/20261015T072909Z-disease-width-readiness/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/recoveries/disease_width_comparison_20261015/compare_ready.py

@@ -0,0 +1,74 @@
+"""CPU-only same-contract disease comparison. Never starts training or generation."""
+from pathlib import Path
+import os,sys,json,csv,time,subprocess,statistics
+R=Path(__file__).resolve().parent;S=R.parents[1];M=S.parents[1];F=M/'plans/paired_finetune_intradataset_20260927';X=M/'plans/f01_trunk4x_complete_20261013'
+sys.path[:0]=[str(S/'tools'),str(S/'recoveries/giroux_width_comparison_20261015')]
+import score_strict as ST
+from validator_fix import install
+from runtime import sha256,atomic_json
+install(ST)
+assert os.environ.get('CUDA_VISIBLE_DEVICES')==''
+COHORTS={'ye2026_sepsis':('adult_sepsis,ped_sepsis','adult_HC,adult_ICU_ctrl,ped_HC,ped_ICU_ctrl'),'imperial_eskd':('covid_positive','covid_negative'),'ji2025_chili':('TOL,FU1','BC,AC')}
+
+def prepare(c,i):
+ rel=Path(c)/'disease'/f'draw{i}';gen=S/'generated'/rel/'lora.npz';sup=ST.read_support(S/'protocol/splits'/rel/'support.json');folder=R/c/f'draw{i}';folder.mkdir(parents=True,exist_ok=True)
+ proof=folder/'BASELINE.json'
+ if proof.exists():
+  b=json.loads(proof.read_text());assert sha256(Path(b['summary']))==b['summary_sha256'];return b
+ bind=ST.validate_generation_arm('lora',gen,sup)
+ old=S/'scores'/rel/'score/summary.json'
+ if old.exists() and json.loads(old.read_text()).get('per_arm',{}).get('lora',{}).get('npz_sha256')==bind['npz_sha256']:source=old
+ else:
+  subprocess.run(['systemctl','--user','is-active','ml-memory-guard.service'],check=True)
+  subprocess.run(['/home/sj_server_1/.local/libexec/ml-memory-guard','--check'],check=True)
+  subprocess.run(['nvidia-smi','--query-compute-apps=pid,process_name,used_memory','--format=csv,noheader'],check=True)
+  available=next(int(x.split()[1])/1048576 for x in Path('/proc/meminfo').read_text().splitlines() if x.startswith('MemAvailable:'))
+  if available<48:raise BlockingIOError('Wait for >=48GiB available before CPU scoring')
+  ids=folder/'test_samples.txt';ids.write_text('\n'.join(sup['test_sample_ids'])+'\n');out=folder/'score';source=out/'summary.json'
+  if source.exists():raise RuntimeError('Unreceipted prior scoring output; inspect')
+  args=[sys.executable,str(F/'tools/score_ft.py'),'--dataset',c,'--out',str(out),'--arm',f'lora={gen}','--condition',f"lora={S/'data'/rel/'test_raw.npz'}",'--donors',str(ids),'--threads','2','--case-groups',COHORTS[c][0],'--control-groups',COHORTS[c][1]]
+  subprocess.run(args,check=True)
+ b={'summary':str(source),'summary_sha256':sha256(source),'binding':bind,'reused_existing_score':source==old};atomic_json(proof,b);return b
+
+def compare(c,i,b):
+ rel=Path(c)/'disease'/f'draw{i}';old=json.loads(Path(b['summary']).read_text());out=X/'payload/paired/strict'/rel;nf=out/'evaluation/score/summary.json';new=json.loads(nf.read_text())
+ a=json.loads((S/'runs'/rel/'lora/RUN.json').read_text());z=json.loads((out/'run/RUN.json').read_text())
+ for k in ['train_rows','inner_val_rows','test_rows_never_loaded_for_training','seed','lr','warmup','batch_size','ema_decay','fixed_steps','selected_step']:assert a[k]==z[k],(c,i,k)
+ for k in ['lora_rank','lora_alpha']:assert a['trainable'][k]==z['trainable'][k]==16
+ assert a['source_checkpoint_sha256']==ST.BASE_SHA
+ assert z['source_checkpoint_sha256']==json.loads((X/'PRETRAIN_VERIFIED.json').read_text())['checkpoint_sha256']
+ g1=json.loads((S/'generated'/rel/'lora.log').read_text());g4=json.loads((out/'generated.log').read_text())
+ for k in ['condition_sha256','seed','steps','weights']:assert g1[k]==g4[k],(c,i,k)
+ import numpy as np
+ with np.load(S/'generated'/rel/'lora.npz',allow_pickle=False) as q,np.load(out/'generated.npz',allow_pickle=False) as t:
+  assert np.array_equal(q['gene_ids'],t['gene_ids']);assert set(q['sample_ids'].astype(str))==set(t['sample_ids'].astype(str))
+ for k in ['contrast_definition','gene_exclusion','deg_definition']:assert old.get(k)==new.get(k),(c,i,k)
+ assert old['donors']==new['donors'];assert old['reference']['reference_sha256']==new['reference']['reference_sha256']
+ assert old['per_arm']['lora']['annotation']['model_pt_sha256']==new['per_arm']['trunk4x']['annotation']['model_pt_sha256']
+ rows=[]
+ for name,j,key in [('x1_LoRA',old,'lora'),('x4_LoRA',new,'trunk4x')]:
+  h=j['headline'][key];rows.append({'cohort':c,'draw':i,'method':name,'composition_l1':h['comp_l1'],'coverage':h['coverage'],'expression_r':h['pb_r_median'],'logfc_r':h['contrast_mean_cp10k_r_gen'],'partial_r':j['per_arm'][key]['contrast']['mean_cp10k']['mean_partial_r'],'deg_overlap':h['deg_top50_sum']})
+ atomic_json(R/c/f'draw{i}/MATCHED.json',{'rows':rows,'same_contract':True,'x1_summary_sha256':b['summary_sha256'],'x4_summary_sha256':sha256(nf),'scope':'Core metrics only; additional statistics may still be running','gpu_operations':0})
+
+def main():
+ while True:
+  states=[];rows=[]
+  for c in COHORTS:
+   for i in range(5):
+    rel=Path(c)/'disease'/f'draw{i}';dest=R/c/f'draw{i}/MATCHED.json';gen=S/'generated'/rel/'lora.npz';nf=X/'payload/paired/strict'/rel/'evaluation/score/summary.json'
+    if dest.exists():rows.extend(json.loads(dest.read_text())['rows']);states.append({'cohort':c,'draw':i,'state':'matched'});continue
+    if not gen.exists():states.append({'cohort':c,'draw':i,'state':'waiting_x1_output_import'});continue
+    try:b=prepare(c,i)
+    except BlockingIOError:states.append({'cohort':c,'draw':i,'state':'waiting_cpu_memory'});continue
+    if not nf.exists():states.append({'cohort':c,'draw':i,'state':'x1_ready_waiting_x4_core_score'});continue
+    compare(c,i,b);rows.extend(json.loads(dest.read_text())['rows']);states.append({'cohort':c,'draw':i,'state':'matched'})
+  if rows:
+   with (R/'COMPARISON.csv').open('w') as f:
+    w=csv.DictWriter(f,fieldnames=rows[0].keys());w.writeheader();w.writerows(rows)
+  atomic_json(R/'STATUS.json',{'state':'complete' if len(rows)==30 else 'waiting_results','matched_draws':len(rows)//2,'total_draws':15,'units':states,'gpu_operations':0})
+  if len(rows)==30:return
+  if '--once' in sys.argv:return
+  time.sleep(120)
+if __name__=='__main__':
+ try:main()
+ except BaseException as e:atomic_json(R/'ERROR.json',{'state':'stopped_needs_inspection','error':str(e)});raise

paper-releases/pbisc-diffusion/updates/20261015T072909Z-disease-width-readiness/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/recoveries/disease_width_comparison_20261015/ji2025_chili/draw1/BASELINE.json

@@ -0,0 +1,18 @@
+{
+  "summary": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/recoveries/disease_width_comparison_20261015/ji2025_chili/draw1/score/summary.json",
+  "summary_sha256": "d4b2661fb04f7927ef27579a3aeea58a9ac114f16ac1303eaaa54f5abeb8db33",
+  "binding": {
+    "arm": "lora",
+    "npz": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/generated/ji2025_chili/disease/draw1/lora.npz",
+    "npz_sha256": "88f760546ae596ec9207d30d463b5912734e877356186833a4752c8925a6bc50",
+    "receipt": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/generated/ji2025_chili/disease/draw1/lora.log",
+    "receipt_sha256": "3c632e6ffb9b94e86b465c0306826aa8f8e2c14b4d66ce391134189b8fb65ef9",
+    "source_binding": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/runs/ji2025_chili/disease/draw1/lora/RUN.json",
+    "source_binding_sha256": "bf59fadc592f4bf0f6d67ae1b53307f3f88d25369b0e2ad417fe153d752d6f8d",
+    "condition_npz": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paper_figures_strictpairs_20261012/data/ji2025_chili/disease/draw1/test_raw.npz",
+    "condition_sha256": "0c873c5de82a9be4c2aba3f1670dddf902c32b351a99677e819367a603fb3bb4",
+    "checkpoint_sha256": "a9a14ff9361e6cc530320810b94100e7a38353374fa54aa3fab7552c8cf4945f",
+    "support_sha256": "4b619899a91698c3840bd1333f6311e2c8fab322cc039b040884fb48acbe15f9"
+  },
+  "reused_existing_score": false
+}