Frank ff52543d8a docs: frank-medium splits weights and KV across both cards, it does not replicate them ...

The loader prints no per-device ledger at this verbosity, so the claim is
measured rather than read: +131072 tokens of context added 4.83 GB to one card
and 4.84 GB to the other (both carry their own attention layers' KV), and
neither card's total usage could hold the 31.46 GB file. Worth writing down
because the retired shape WAS two full copies — that is what a replica is, and
it is why the 29.8 GB FP8 checkpoint could never run there.

2026-09-16 15:52:18 -07:00

..

.obsidian

docs: rename docs/ -> serverAppConfigs-docs/ per system-wide PROJECTNAME-docs convention

2026-08-28 02:20:06 -07:00

Architecture

docs: frank-medium runbook, fleet topology and client cutover

2026-09-15 19:27:04 -07:00

Dankclaw

docs: three dangling links, and a gate so they cannot accumulate

2026-09-16 07:43:33 -07:00

Dankserver

docs: frank-medium runbook, fleet topology and client cutover

2026-09-15 19:27:04 -07:00

Infrastructure

frank-medium: serve the whole 262144-token window, and repoint every signal to llama.cpp

2026-09-16 15:29:40 -07:00

LLM

docs: frank-medium splits weights and KV across both cards, it does not replicate them

2026-09-16 15:52:18 -07:00

Research

docs: rename frank-default-large/small/sightsound to frank-large/small/npu across vault

2026-09-03 16:17:14 -07:00

index.md

docs: frank-medium is llama.cpp with DFlash2 on both cards

2026-09-16 15:15:02 -07:00

log.md

docs: the vault and AGENTS.md describe two Dankclaw endpoints

2026-09-16 14:52:51 -07:00