v0.6.9.3
AMD/OpenCL: --force-model and --tier now work (fixes #7). The AMD build previously pinned the light tier (Gemma-3-4B) and silently ignored all tier flags. Now:
--force-model <name>selects the mining model on AMD. It is process-wide on AMD (one resident model per process): the first entry applies to all cards. Example:--force-model default→ Dolphin-8B.--tier <name>/--light/--highetc. are also honored (process-wide).- No flags = unchanged (light tier, the OOM-safe default).
All platforms: --force-model now bypasses the VRAM capability gate, as documented — the forced model loads regardless of VRAM fit. A card too small for the forced model will fail to load it (power-user knob; the log says so clearly).
Version banner/user-agent now reports 0.6.9 (was 0.6.8 — cosmetic; several users asked).
Package versions: CUSTOM_VERSION 0.6.9.3 / BUILD 22. Windows zips are attached by CI. AMD Linux packages follow separately.