• Joined on 2026-01-29
mAi deleted branch mai/icarus/upc-formalities-table from m/paliad 2026-07-29 10:41:15 +00:00
mAi deleted branch mai/icarus/cron-guards-mute from m/paliad 2026-07-29 10:41:14 +00:00
mAi deleted branch mai/linus/rls-s6-the-production from m/paliad 2026-07-29 10:41:13 +00:00
mAi deleted branch mai/linus/hlcpat-css-is-a-third from m/paliad 2026-07-29 10:41:13 +00:00
mAi deleted branch mai/pike/re-vendor-patentstyle from m/paliad 2026-07-29 10:41:12 +00:00
mAi deleted branch mai/knuth/give-paliad-its-own from m/paliad 2026-07-29 10:41:11 +00:00
mAi deleted branch mai/ritchie/regenerate-offices from m/paliad 2026-07-29 10:41:11 +00:00
mAi deleted branch mai/knuth/classifydrift-case-a from m/paliad 2026-07-29 10:41:10 +00:00
mAi deleted branch mai/ford/otto-502-firm-rename-the from m/paliad 2026-07-29 10:41:09 +00:00
mAi deleted branch mai/ford/internal-models-models from m/paliad 2026-07-29 10:41:09 +00:00
mAi deleted branch mai/knuth/after-mig-206-deploys from m/paliad 2026-07-29 10:41:09 +00:00
mAi pushed to main at m/paliad 2026-07-29 10:34:59 +00:00
2697a4b021 Merge branch 'mai/knuth/after-mig-206-deploys' (seed formalities_type_values; both fixtures moved together)
2eb7bfcf91 test(fixture): seed formalities_type_values, and move both fixtures together
Compare 2 commits »
mAi opened issue m/mGPUmanager#13 2026-07-29 10:34:34 +00:00
The broker can reclaim VRAM it did not allocate, but nothing stops a direct client taking it — a bare ollama run still OOM-kills whisper
mAi pushed to mai/knuth/after-mig-206-deploys at m/paliad 2026-07-29 10:33:11 +00:00
2eb7bfcf91 test(fixture): seed formalities_type_values, and move both fixtures together
mAi created branch mai/knuth/after-mig-206-deploys in m/paliad 2026-07-29 10:33:11 +00:00
mAi commented on issue m/mGPUmanager#6 2026-07-29 10:33:07 +00:00
Broker must be the SOLE VRAM authority — close bypass gaps so out-of-band ollama holds get evicted (imagen starved by 24h keep-alive)

Partly closed — reclaim works, prevention does not

What now works, and is deployed on mRock:

Per-consumer residency comes from the GPU instead of from the consumers' own health responses,…

mAi pushed to main at m/paliad 2026-07-29 10:23:06 +00:00
489b99eae7 Merge branch 'mai/carmack/formalities-submission' (mig 206: per-value provenance on formalities_submission_types + label_en correction)
4f93464499 test(fixture): re-take reference-seed with the formalities catalogs (21, was 19)
232fc6d39f test(fixture): seed the formalities catalog tables — mig 206's FK has nothing to point at without them
bbd1d5d9cd Merge main (reference-seed refreshed to match the 205 snapshot — 5dcd060)
1d1e630121 Merge main (snapshot refreshed to prod HEAD 205 — 39f2319) into mig 206 branch
Compare 8 commits »
mAi pushed to mai/carmack/formalities-submission at m/paliad 2026-07-29 10:20:28 +00:00
4f93464499 test(fixture): re-take reference-seed with the formalities catalogs (21, was 19)
mAi commented on issue m/mGPUmanager#4 2026-07-29 10:19:59 +00:00
Broker reports per-consumer gpu_resident_mib=0 for externally-started consumers → eviction finds "no candidates" and never reclaims their VRAM

Deployed to mRock from main 531cc3b and verified live.

/v1/status now reports measured per-consumer residency, matching nvidia-smi exactly:

comfyui        6362   (nvidia-smi pid 2800: …
mAi commented on issue m/mGPUmanager#12 2026-07-29 10:19:46 +00:00
comfyui's VRAM need is not a constant: it varies cold-vs-warm (8410 vs 6170 MiB) and with the requested resolution

Why measuring harder will not answer this

This issue argues a single constant cannot express a requirement that varies by cold/warm and resolution. The measurements say something stronger, and…