Archive 2 checkpoints (1-2)
Browse files
paper-releases/pbisc-diffusion/updates/20261005T142225Z-rare-states-scratch-vs-pretrained/diffusion/metadata/NOTE.md
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# PBISC-diffusion update 20261005T142225Z-rare-states-scratch-vs-pretrained
|
| 2 |
+
|
| 3 |
+
Rare states, from scratch vs pretrained (plans/rare_states_screen_20261004, 2026-10-05): new --scope loco in tools/sepsis_states.py for the cross-cohort arms with all 82 sepsis donors held out (scratch milestones 50k-162k, LoRA r16/r64, full FT, xcal; own-cohort LoRA and zero-shot as references), sequential extracts (tools/run_loco_extract.sh; every NPZ streamed from spark-0110 sha-checked, benchmark Poisson read-out reproduced), outputs out/sepsis_loco_*; the fold-0 within-cohort scratch arms; a donor-bootstrap / confound-audit / two-pass verification analysis (out/scratch_vs_lora_20261005/, REPORT.md). No new generation; no inference locations.
|
| 4 |
+
|
| 5 |
+
Payload states are recorded as supplied. Inference locations are pointers, not archived generated outputs.
|
paper-releases/pbisc-diffusion/updates/20261005T142225Z-rare-states-scratch-vs-pretrained/diffusion/metadata/inference_locations.json
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"locations": [],
|
| 3 |
+
"project_id": "pbisc-diffusion",
|
| 4 |
+
"update_id": "20261005T142225Z-rare-states-scratch-vs-pretrained"
|
| 5 |
+
}
|