sigilant-runner · Phi-3.5-mini-instruct · L4 · vllm · 4 configs

Config | TPS | TPS p95 | TTFT | TTFT p95 | ITL | PPL | TPS% | TTFT% | PPL% | Score
FP16_BASELINE · ctx:16384 · kv:k16v16 · long  <- best | 47.4 | n/a | 643.9 | n/a | 21.18 | 3.06 | 99.8 | 100.0 | 99.0 | 99
FP16_BASELINE · ctx:8192 · kv:k16v16 · default | 47.4 | n/a | 1360.3 | n/a | 21.17 | 3.03 | 99.8 | 47.3 | 100.0 | 89
FP16_BASELINE · ctx:16384 · kv:k8v8 · default | 47.5 | n/a | 1383.0 | n/a | 21.14 | 3.07 | 100.0 | 46.6 | 98.7 | 89
FP16_BASELINE · ctx:32768 · kv:k8v8 · long | 47.4 | n/a | 1406.7 | n/a | 21.18 | 3.06 | 99.8 | 45.8 | 99.0 | 89

Best config: FP16_BASELINE · ctx:16384 · kv:k16v16 · long
Auto baseline compare: score Δ=0.00 TPS Δ=0.00 TTFT Δ=0.0ms PPL Δ=0.00
Confidence: target=medium gap_before=10.10% var_before=n/a% replay=False(disabled) gap_after=10.10%

PPL is a quality proxy, not production validation.
Full production safety and long-context certification require Sigilant Optimizer.
