sglang/ GB200 perf

dsr1-fp4-1k1k-mid-curve

concurrency 2,048 · 8 days ago

cron
passed
re-run
Commit

Reenable MNNVL backend for FlashInfer allreduce fusion (#23402)

wenscarl·8 days ago
PR #23402 · Reenable MNNVL backend for FlashInfer allreduce fusion
Run
GitHub Actions27591824405-1
Slurm job5312
GPUs48·prefill 16 / decode 32
ISL / OSL1024 / 1024

Metrics

28 captured

best_of

1.00

burstiness

1.00

completed

20,480

duration

213

max_concurrency

2,048

mean_e2el_ms

20,521ms

mean_itl_ms

998ms

mean_tpot_ms

20.36ms

mean_ttft_ms

1,759ms

median_e2el_ms

20,074ms

median_itl_ms

1,034ms

median_tpot_ms

20.90ms

median_ttft_ms

819ms

num_prompts

20,480

output_throughput

88,739tok/s

p99_e2el_ms

32,814ms

p99_itl_ms

1,532ms

p99_tpot_ms

21.59ms

p99_ttft_ms

13,739ms

peak_output_tokens_per_s

105,935s

request_throughput

96.21tok/s

std_e2el_ms

3,063ms

std_itl_ms

209ms

std_tpot_ms

1.32ms

std_ttft_ms

2,798ms

total_input_tokens

18,880,692

total_output_tokens

18,888,974

total_token_throughput

177,440tok/s