One number per rank: FRANK_ARCHON_HOST_POOL_MIB 44,928 -> 45,568, and the TTM floor with it (150,000 -> 152,000).
Why
The 1.8.1 boot resolved the cache to 556,348 tokens = 2.8 sessions at 200,000. The capacity leg of scripts/probe-frank-archon-kv.py came back 400 Bad Request in 0.0 s for all three concurrent full-height requests. The endpoint owes 3 full-height conversations at once; the cutover broke that silently — /health, /v1/models and every shorter conversation answer fine.
The budget resolver gives the expert plane whatever the pinned host pool cannot hold and the cache the rest (core/mem/vram_budget.h: "the cache gets the rest up to its ceiling"). So the pool is the lever. Tokens scale with the GiB the cache wins (~8.1 KiB/token at fp8): +640 MiB a rank is ~ +0.63 GiB ~ +44,000 tokens -> 600,000 with margin, at ~2.5 % off the expert slab.
Evidence
The boot ledger and the capacity leg decide, not the arithmetic: the kv: line must read >= 600,000 tokens / 3.0 sessions and leg B must answer all three markers. If the slab reduction moves prefill, the slab line comes off the same ledger.
## What
One number per rank: `FRANK_ARCHON_HOST_POOL_MIB` 44,928 -> 45,568, and the TTM floor with it (150,000 -> 152,000).
## Why
The 1.8.1 boot resolved the cache to **556,348 tokens = 2.8 sessions** at 200,000. The capacity leg of `scripts/probe-frank-archon-kv.py` came back `400 Bad Request` in 0.0 s for all three concurrent full-height requests. The endpoint owes 3 full-height conversations at once; the cutover broke that silently — `/health`, `/v1/models` and every shorter conversation answer fine.
The budget resolver gives the expert plane whatever the pinned host pool cannot hold and the cache the rest (`core/mem/vram_budget.h`: "the cache gets the rest up to its ceiling"). So the pool is the lever. Tokens scale with the GiB the cache wins (~8.1 KiB/token at fp8): +640 MiB a rank is ~ +0.63 GiB ~ +44,000 tokens -> 600,000 with margin, at ~2.5 % off the expert slab.
## Evidence
The boot ledger and the capacity leg decide, not the arithmetic: the `kv:` line must read >= 600,000 tokens / 3.0 sessions and leg B must answer all three markers. If the slab reduction moves prefill, the slab line comes off the same ledger.
---
cwd: /home/blank/.omp/wt/wt-archon-181b · session: wt-20261019-020309-6d1fc1f
The 1.8.1 boot resolved the budget resolver to 4.20 GiB of cache a card —
556,348 tokens, 2.8 sessions at full height — so three concurrent
full-height requests are refused with a 400 at admission: the endpoint's
3-slot contract is broken by the cutover, silently, because /health and
every shorter conversation still answer.
The resolver gives the expert plane whatever the pinned host pool cannot
hold and the cache the rest, so the pool is the lever: +640 MiB a rank
moves that much expert demand into host RAM and the cache wins it back.
Tokens scale with the GiB won (8.11 KiB/token at fp8), so ~0.63 GiB buys
~44,000 tokens — 600,000 with margin — at ~2.5% off the expert slab.
TTM floor moves with the pool (2 x pool + 48,824 + prefix + MNBT =
150,200 MiB computed by the ledger test).
Verified by the boot ledger's `kv:` line and the probe's capacity leg,
not by this arithmetic.
dank
merged commit 64765c6a47 into main2026-10-19 10:30:13 -07:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
What
One number per rank:
FRANK_ARCHON_HOST_POOL_MIB44,928 -> 45,568, and the TTM floor with it (150,000 -> 152,000).Why
The 1.8.1 boot resolved the cache to 556,348 tokens = 2.8 sessions at 200,000. The capacity leg of
scripts/probe-frank-archon-kv.pycame back400 Bad Requestin 0.0 s for all three concurrent full-height requests. The endpoint owes 3 full-height conversations at once; the cutover broke that silently —/health,/v1/modelsand every shorter conversation answer fine.The budget resolver gives the expert plane whatever the pinned host pool cannot hold and the cache the rest (
core/mem/vram_budget.h: "the cache gets the rest up to its ceiling"). So the pool is the lever. Tokens scale with the GiB the cache wins (~8.1 KiB/token at fp8): +640 MiB a rank is ~ +0.63 GiB ~ +44,000 tokens -> 600,000 with margin, at ~2.5 % off the expert slab.Evidence
The boot ledger and the capacity leg decide, not the arithmetic: the
kv:line must read >= 600,000 tokens / 3.0 sessions and leg B must answer all three markers. If the slab reduction moves prefill, the slab line comes off the same ledger.cwd: /home/blank/.omp/wt/wt-archon-181b · session: wt-20261019-020309-6d1fc1f