qqtang/sskj
1
2
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

enable initial backoff for all add16 benchmarks

Browse Source
This commit is contained in:
SSKJ Dev 2026-07-16 09:05:37 +00:00
parent c2a3024a1b
commit 589d3b079d
6 changed files with 18 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
experiments/h20/dsv4_h20_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh
View File

@ -180,5 +180,8 @@ export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUE
export SEARCH_START_CONCURRENCY=16 export SEARCH_START_CONCURRENCY=16
export SEARCH_ADDEND=16 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"
adaptive_main "$@" adaptive_main "$@"

3
experiments/h200/dsv4_h200_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh
View File

@ -180,5 +180,8 @@ export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUE
export SEARCH_START_CONCURRENCY=16 export SEARCH_START_CONCURRENCY=16
export SEARCH_ADDEND=16 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"
adaptive_main "$@" adaptive_main "$@"

3
experiments/h200/dsv4_h200_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh
View File

@ -163,5 +163,8 @@ export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUE
export SEARCH_START_CONCURRENCY=16 export SEARCH_START_CONCURRENCY=16
export SEARCH_ADDEND=16 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"
adaptive_main "$@" adaptive_main "$@"

3
experiments/p800/dsv4_p800_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh
View File

@ -148,5 +148,8 @@ export BENCH_DATASET_NAME CONTAINER_DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP
export SEARCH_START_CONCURRENCY=16 export SEARCH_START_CONCURRENCY=16
export SEARCH_ADDEND=16 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"
adaptive_main "$@" adaptive_main "$@"

3
experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh
View File

@ -180,5 +180,8 @@ export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUE
export SEARCH_START_CONCURRENCY=16 export SEARCH_START_CONCURRENCY=16
export SEARCH_ADDEND=16 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"
adaptive_main "$@" adaptive_main "$@"

3
experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh
View File

@ -163,5 +163,8 @@ export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUE
export SEARCH_START_CONCURRENCY=16 export SEARCH_START_CONCURRENCY=16
export SEARCH_ADDEND=16 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"
adaptive_main "$@" adaptive_main "$@"