sigilant-sweep · 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 | 58.3 | n/a | 141.3 | n/a | 17.22 | 2.68 | 98.0 | 100.0 | 100.0 | 99
FP16_BASELINE · ctx:16384 · kv:k8v8 · default | 58.4 | n/a | 736.1 | n/a | 17.18 | 2.68 | 98.2 | 19.2 | 100.0 | 83
FP16_BASELINE · ctx:8192 · kv:k16v16 · default | 59.5 | n/a | 903.4 | n/a | 16.95 | 2.71 | 100.0 | 15.6 | 98.9 | 83
FP16_BASELINE · ctx:32768 · kv:k8v8 · long | 43.6 | n/a | 1297.4 | n/a | 23.03 | 2.70 | 73.3 | 10.9 | 99.3 | 71

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=16.16% var_before=n/a% replay=False(disabled) gap_after=16.16%

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