diff --git a/experiments/h20/dsv4_h20_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh b/experiments/h20/dsv4_h20_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh index fe1c162..c98b531 100755 --- a/experiments/h20/dsv4_h20_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh +++ b/experiments/h20/dsv4_h20_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh @@ -183,5 +183,8 @@ export SEARCH_ADDEND=16 # If the initial concurrency violates the TTFT SLO, search downward. Stop at # the first acceptable value (16 -> 8; only try 1 when 8 still violates it). export SEARCH_INITIAL_BACKOFF_CONCURRENCIES="8 1" +# When concurrency 1 still has a severely excessive TTFT, stop the remaining +# shapes in this TP/DP group. Zero disables this rule. +export TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-8000}" adaptive_main "$@" diff --git a/experiments/h200/dsv4_h200_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh b/experiments/h200/dsv4_h200_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh index 29a7eca..375fec7 100755 --- a/experiments/h200/dsv4_h200_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh +++ b/experiments/h200/dsv4_h200_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh @@ -183,5 +183,8 @@ export SEARCH_ADDEND=16 # If the initial concurrency violates the TTFT SLO, search downward. Stop at # the first acceptable value (16 -> 8; only try 1 when 8 still violates it). export SEARCH_INITIAL_BACKOFF_CONCURRENCIES="8 1" +# When concurrency 1 still has a severely excessive TTFT, stop the remaining +# shapes in this TP/DP group. Zero disables this rule. +export TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-8000}" adaptive_main "$@" diff --git a/experiments/h200/dsv4_h200_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh b/experiments/h200/dsv4_h200_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh index a30729e..1b2b1b2 100755 --- a/experiments/h200/dsv4_h200_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh +++ b/experiments/h200/dsv4_h200_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh @@ -166,5 +166,8 @@ export SEARCH_ADDEND=16 # If the initial concurrency violates the TTFT SLO, search downward. Stop at # the first acceptable value (16 -> 8; only try 1 when 8 still violates it). export SEARCH_INITIAL_BACKOFF_CONCURRENCIES="8 1" +# When concurrency 1 still has a severely excessive TTFT, stop the remaining +# shapes in this TP/DP group. Zero disables this rule. +export TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-8000}" adaptive_main "$@" diff --git a/experiments/p800/dsv4_p800_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh b/experiments/p800/dsv4_p800_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh index 4e8e30c..91176a6 100755 --- a/experiments/p800/dsv4_p800_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh +++ b/experiments/p800/dsv4_p800_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh @@ -151,5 +151,8 @@ export SEARCH_ADDEND=16 # If the initial concurrency violates the TTFT SLO, search downward. Stop at # the first acceptable value (16 -> 8; only try 1 when 8 still violates it). export SEARCH_INITIAL_BACKOFF_CONCURRENCIES="8 1" +# When concurrency 1 still has a severely excessive TTFT, stop the remaining +# shapes in this TP/DP group. Zero disables this rule. +export TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-8000}" adaptive_main "$@" diff --git a/experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh b/experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh index 7876ff9..50f0b6b 100755 --- a/experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh +++ b/experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh @@ -183,5 +183,8 @@ export SEARCH_ADDEND=16 # If the initial concurrency violates the TTFT SLO, search downward. Stop at # the first acceptable value (16 -> 8; only try 1 when 8 still violates it). export SEARCH_INITIAL_BACKOFF_CONCURRENCIES="8 1" +# When concurrency 1 still has a severely excessive TTFT, stop the remaining +# shapes in this TP/DP group. Zero disables this rule. +export TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-8000}" adaptive_main "$@" diff --git a/experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh b/experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh index 8912882..aee8e20 100755 --- a/experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh +++ b/experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh @@ -166,5 +166,8 @@ export SEARCH_ADDEND=16 # If the initial concurrency violates the TTFT SLO, search downward. Stop at # the first acceptable value (16 -> 8; only try 1 when 8 still violates it). export SEARCH_INITIAL_BACKOFF_CONCURRENCIES="8 1" +# When concurrency 1 still has a severely excessive TTFT, stop the remaining +# shapes in this TP/DP group. Zero disables this rule. +export TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-8000}" adaptive_main "$@"