diff --git a/experiments/h20/glm_h20_vllm_tp_dp_matrix/adaptive_config.env b/experiments/h20/glm_h20_vllm_tp_dp_matrix/adaptive_config.env index d725207..5c8d5ea 100644 --- a/experiments/h20/glm_h20_vllm_tp_dp_matrix/adaptive_config.env +++ b/experiments/h20/glm_h20_vllm_tp_dp_matrix/adaptive_config.env @@ -21,7 +21,7 @@ PLATEAU_PATIENCE="${PLATEAU_PATIENCE:-2}" # TTFT SLO early-stop settings. # When ttft_p95_ms exceeds TTFT_SLO_MS, stop searching the current (ISL, OSL) # shape and move on to the next scenario. -TTFT_SLO_MS="${TTFT_SLO_MS:-4000}" +TTFT_SLO_MS="${TTFT_SLO_MS:-6000}" ENABLE_TTFT_SLO_STOP="${ENABLE_TTFT_SLO_STOP:-1}" # Keep the same random workload semantics as the fixed matrix baseline. diff --git a/experiments/h20/glm_h20_vllm_tp_dp_matrix/compare.py b/experiments/h20/glm_h20_vllm_tp_dp_matrix/compare.py index 002fa0c..2dd95da 100644 --- a/experiments/h20/glm_h20_vllm_tp_dp_matrix/compare.py +++ b/experiments/h20/glm_h20_vllm_tp_dp_matrix/compare.py @@ -18,7 +18,7 @@ def load_result(result_root: Path) -> dict: def slo_status(ttft_p95_ms: float, tpot_mean_ms: float, - ttft_limit_ms: float = 3000.0, tpot_limit_ms: float = 50.0) -> str: + ttft_limit_ms: float = 5000.0, tpot_limit_ms: float = 50.0) -> str: ttft_ok = ttft_p95_ms < ttft_limit_ms tpot_ok = tpot_mean_ms < tpot_limit_ms if ttft_ok and tpot_ok: