ark
Star
Fork
Watch

Commits on buer
buer: bring the 1 GB headscale VPS into ark
committed
focalor: strata's vision encoder runs on the RTX 4060 (CUDA build for sm_89)
committed
focalor: strata no longer needs /strata-fast mounted
committed
focalor: strata back on Qwen3.8-Flash-Next GSQ-RCO IQ3_XXS with niri's think1, 200K context, served as deepseek-v4.1-flash
committed
tangled: master 9e6f01a
committed
trimounts: knot blob store
committed
trimounts: knot on tangled core master (1.0), admission closed
committed
valefar: pia-exit tries every DC ovpnudp server, not one pinned address
committed
valefar: spindle allows cross-origin reads
committed
valefar: spindle on tangled 1.0, served as did:web:ci.klbr.net
committed
valefar: swap to zram first, the NVMe partition only as overflow
committed
valefar: spindle metrics on the tailnet address for victoriametrics
committed
focalor: strata keeps and splices conversations with pictures (niri back after another request: 94 s -> 0.4 s), slot folder 16 GB
committed
focalor: nvidia-smi patched with patchelf 0.15 (the driver's copy segfaults, NixOS/patchelf#152)
committed
focalor: strata streams an empty delta every 15 s while reading a long prompt (niri's harness idle timeout ignores SSE comments)
committed
focalor: strata context 320K (niri's history is 224K tokens)
committed
focalor: strata DeepSeek-V4.1 sees images (vision tower on the 4060)
committed
focalor: strata keeps the 4060 expert GPU and its cache (3b5d464 dropped them by mistake)
committed
focalor: strata Cache-Prior over the 5070 and 4060 only, swaps at window ticks (63K decode 10.9 -> 12.0 tok/s, 16K 12.7 -> 13.5, deterministic)
committed
focalor: strata Cache-Prior favours RAM-store and 4060 experts too (63K decode 10.0 -> 10.9 tok/s, misses -40%, KL no worse)
committed
focalor: strata 4060 experts follow the conversation (63K-context decode 9.0 -> 10.0 tok/s, 16K 10.9 -> 12.3)
committed
focalor: strata SCR keeps conversation slots (a compaction after another conversation 79 -> 16 s, restart restores from disk)
committed
focalor: strata reuses edited prompts (SCR: 63K compaction 77 -> 16.5 s, continuation KL ~0.015), exits on QUIT at once
committed
focalor: strata Cache-Prior 0.15 (perplexity-neutral), the 4060 as expert GPU, faster prompts (63.5K cold 116 -> 70 s); prose 13.7, code 21.3 tok/s
committed
focalor: strata suffix drafts up to 7 tokens (code 20.4 -> 22.2 tok/s, exact), store keeps failed pinned slots for their last holder
committed
focalor: strata DeepSeek adds in-place store reads, top-3 read-ahead and suffix drafting (code rewrites 12.5 -> 20.4 tok/s, exact)
committed
focalor: strata serves DeepSeek-V4.1-Flash (ISTA GSQ-RCO IQ2_XXS) on its own engine, AVX2 build, 200K context, Cache-Prior 0.3
committed
chernobog: build udiskie without its keyring tests chernobog-openssh
committed
nodejs-security: patched node as separate attributes, not a global replacement
committed
chernobog: sign the hand-installed kernel modules with kmodigest
committed
valefar: bonsai's NVMe prompt cache tier is 20 GB
committed
chernobog: drop the depmod index files v4l2loopback 0.15.4 ships
committed
chernobog: drop meowtd
committed
dzwonek: rate-limit headscale's unauthenticated paths, cap its memory
committed
chernobog: OpenSSH for mayer instead of Tailscale SSH
committed
valefar: the nixos-large CI image gets 12 vCPUs
committed
valefar: spindle microVM overlays on the NVMe
committed
valefar: spindle microVM overlays on a sync=disabled dataset
committed
valefar: a 12 GiB / 8 vCPU nixos-large CI image, bonsai keeps 8 GB of prompt cache in RAM, qBittorrent's page cache stops at 2 GB
committed
focalor: strata serves niri on Swift 1.5 GSQ-RCO IQ3_S (held-out NLL 2.394 -> 2.344, decode 65 -> 51 t/s)
committed
focalor: strata decodes without the routing-aware lookahead (+3% decode)
committed
focalor: strata skips GPU-missed experts ranked 5th or lower (decode +10%)
committed
valefar: pia-exit tunnel back to OpenVPN UDP
committed
valefar: cap the CPU at 3.8 GHz instead of 4.4
committed
focalor: strata serves niri's think1 merged into Swift 1.5 (median thinking 924 -> 646 characters)
committed
focalor: 128 KiB read-ahead on the strata array
committed
focalor: drop the depmod index files v4l2loopback 0.15.4 ships, which clash with the kernel's
committed
focalor: strata reads from a RAID0 over both NVMes, /nix moves to the SN7100, zram swap
committed
valefar: nightly encrypted offsite backup to GCS
committed
dzwonek: lisya may use exit nodes too
committed
valefar: pia-exit clamps MSS and lets tailscaled's UDP out eth0
committed
nodejs: October 2026 security release (Brotli RCE CVE-2026-111497)
committed
valefar: nvidia-smi patched with patchelf 0.15, which keeps .dynamic writable
committed
improve deploy script
committed
valefar: fetch searxng with fetchFromGitHub
committed
fix js service ca trust
committed
disable inverter, block emudhra ca
committed
chore(nix): update inputs [skip ci]
committed
awa
committed
focalor: niri matches the MSI monitor on the RTX 4060 too
committed