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 · ctx:32768 · kv:k8v8 · long  <- best | 57.0 | 57.0 | 898.0 | 898.6 | 14.09 | 3.02 | 99.8 | 100.0 | 100.0 | 100
FP16 · ctx:8192 · kv:k16v16 · default | 57.1 | 57.1 | 897.2 | 899.0 | 14.07 | 3.06 | 100.0 | 100.0 | 98.7 | 99
FP16 · ctx:16384 · kv:k8v8 · default | 57.0 | 57.1 | 897.5 | 898.9 | 14.08 | 3.06 | 99.9 | 100.0 | 98.7 | 99
FP16 · ctx:16384 · kv:k16v16 · long | 57.0 | 57.1 | 897.8 | 921.1 | 14.08 | 3.06 | 99.9 | 98.7 | 98.7 | 99

Best config: FP16 · ctx:32768 · kv:k8v8 · long
Auto baseline compare: score Δ=0.00 TPS Δ=0.00 TTFT Δ=0.0ms PPL Δ=0.00
                     TPS p95 Δ=0.00 TTFT p95 Δ=0.0ms
Confidence: target=medium gap_before=1.00% var_before=0.07% replay=False(disabled) gap_after=1.00%

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