-
Notifications
You must be signed in to change notification settings - Fork 393
Pull requests: osaurus-ai/osaurus
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Refuse background model loads inside the actor, not before it
bug
Something isn't working
#1994
opened Jul 12, 2026 by
jjang-ai
Contributor
Loading…
Add opt-in materialized weight loading override
#1993
opened Jul 12, 2026 by
k1x0r
Loading…
2 of 9 tasks
Surface engine-path attribution and predicted cache topology through diagnostics
#1962
opened Jul 10, 2026 by
mimeding
Contributor
Loading…
Download large model shards as parallel HTTP Range requests
#1960
opened Jul 10, 2026 by
jjang-ai
Contributor
Loading…
Harden #1903's load-time MTP warmup: drain deadline, gate-busy skip, lease pinning
#1929
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add cross-model bandwidth budgeter for multi-model big-RAM serving (flag-off)
#1926
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Guardrail retry: recover from pre-content degeneration with safe settings
#1925
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add
bench --calibrate memory-bandwidth probe + decode estimates in show
#1923
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add live stream guardrails: degeneration abort + template-leak detection
#1922
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Session-freeze composer inputs on /agents/{id}/run for KV prefix reuse
#1921
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add opt-in auto batch sizing driven by observed demand, tier-capped
#1919
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Micro-batch local embeddings into one batched MLX forward
#1917
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add
osaurus bench --gauntlet: measured model-capability probes writing the ledger
#1916
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add thermal/chassis awareness: ChipProfile v2 + ThermalStateMonitor + first consumer
#1915
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Add
osaurus bench --kv-matrix: automated fp16-vs-TurboQuant KV codec evidence
#1914
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Tier the MLX buffer-cache cap by RAM (RAM/10 small, RAM/6 capped 24 GiB big)
#1911
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Derive idle-residency default from RAM tier (2 min / 15 min / resident)
#1902
opened Jul 7, 2026 by
mimeding
Contributor
Loading…
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.