Run frontier MoE models on hardware you already own β pure C, zero deps, experts streamed from disk. Tiny engine, immense model. π¦
C
docs: st_dev volume identity is best-effort; document the new veto reasons (#379)
monotophic7e7f0a6f7210
diff --git a/docs/ENVIRONMENT.md b/docs/ENVIRONMENT.mdindex bca64141..48a51c59 100644--- a/docs/ENVIRONMENT.md+++ b/docs/ENVIRONMENT.md@@ -92,17 +92,26 @@ re-measuring. Details that matter when you meet this file in the wild:- **Contamination veto.** If the shard offers fewer than 64 MB of such coldwindows, the measurement is refused: nothing is cached, one stderr lineexplains the deferral, the conservative (slow-storage) defaults hold, and- the probe simply retries on the next, colder, startup.+ the probe simply retries on the next, colder, startup. The same veto (with+ its own honest message) fires for an under-allocated shard β a sparse or+ still-downloading file whose "cold" pages are holes that would measure as+ RAM-speed zero-fill β and for a shard too small to ever offer 64 MB of+ probe windows. The probe measures the largest `.safetensors` in the dir.- **Format (v2).** One line, `v2 <gbs> <st_dev>` β the measured GB/s and the`st_dev` of the model dir's volume at measurement time. The grammar isstrict (plain digits, `0 < gbs < 1000`; no inf/nan/hex/exponents) and bothreaders β the C engine and `coli doctor`/`coli plan` β accept exactly thesame bytes; anything else is ignored and re-probed, never trusted.-- **Volume identity.** The cache is honored only while its recorded `st_dev`- matches the model dir's current volume. Copy or rsync the model dir- (including this hidden file) to another drive and the engine re-probes there- instead of inheriting the old drive's number; doctor/plan likewise stop- showing the stale value.+- **Volume identity (best-effort).** The cache is honored only while its+ recorded `st_dev` matches the model dir's current volume, so copying or+ rsyncing the model dir (including this hidden file) to another drive+ normally triggers a re-probe there instead of inheriting the old drive's+ number; doctor/plan likewise stop showing the stale value. This is+ best-effort, not an identity guarantee: macOS recycles `st_dev` values, so+ a cache carried to an external volume that happens to be assigned the old+ device id (e.g. drives attached one after another in the same slot) will be+ wrongly trusted until deleted. When in doubt after moving a model dir,+ delete `.coli_ssd`. True volume-UUID identity is a named follow-up.- **Legacy upgrade.** A pre-v2 bare-number cache (written before steeringexisted, so possibly warm-contaminated) is re-measured once on the nextstartup and rewritten as v2.