Files
serverAppConfigs/tests
Frank 8d88b3a0a9 frank-small: int8 KV (per-token, per-head) — doubles the pool
fp8 KV is hardware-dead on the 3090 (sm_86, no cvt.e4m3x2), so int8 is
the cheapest quant the TRITON_ATTN backend can run: pure ALU dequant,
same backend, no fp8 tensor cores needed. 289,647 -> ~579k-token pool,
two -> four full-context sessions. Quality verified against the bf16
baseline (needle-in-haystack at 213k + short battery) before the swap.
2026-09-18 23:24:37 -07:00
..