o áÙ·i¯ ã@s@ddlZddlmZddlmZddlmZGdd„deƒZdS)éN)Ú add_cli_args©Úmain)ÚOmniBenchmarkSubcommandBasec@sHeZdZdZdZdZedejddfdd„ƒZ e d ej ddfd d „ƒZ dS) ÚOmniBenchmarkServingSubcommandz&The `serve` subcommand for vllm bench.Úservez(Benchmark the online serving throughput.ÚparserÚreturnNcCsHt|ƒ|jD]}|jdkrd|_|jdkrd|_|jdkr!d|_qdS)NÚpercentile_metricszÌComma-separated list of selected metrics to report percentiles.This argument specifies the metrics to report percentiles.Allowed metric names are "ttft", "tpot", "itl", "e2el", "audio_ttfp", "audio_rtf". Úrandom_mm_limit_mm_per_prompta Per-modality hard caps for items attached per request, e.g. '{"image": 3, "video": 0, "audio": 1}'. The sampled per-request item count is clamped to the sum of these limits. When a modality reaches its cap, its buckets are excluded and probabilities are renormalized.Úrandom_mm_bucket_configa[The bucket config is a dictionary mapping a multimodal itemsampling configuration to a probability.Currently allows for 3 modalities: audio, images and videos. A bucket key is a tuple of (height, width, num_frames)The value is the probability of sampling that specific item. Example: --random-mm-bucket-config {(256, 256, 1): 0.5, (720, 1280, 16): 0.4, (0, 1, 5): 0.10} First item: images with resolution 256x256 w.p. 0.5Second item: videos with resolution 720x1280 and 16 frames Third item: audios with 1s duration and 5 channels w.p. 0.1OBS.: If the probabilities do not sum to 1, they are normalized.)rÚ_actionsÚdestÚhelp)ÚclsrÚaction©rú]/home/ubuntu/.local/lib/python3.10/site-packages/vllm_omni/entrypoints/cli/benchmark/serve.pyrs  ÿ ÿ ÿ€ðz+OmniBenchmarkServingSubcommand.add_cli_argsÚargscCs t|ƒdS)Nr)rrrrÚcmd1s z"OmniBenchmarkServingSubcommand.cmd) Ú__name__Ú __module__Ú __qualname__Ú__doc__ÚnamerÚ classmethodÚargparseÚArgumentParserrÚ staticmethodÚ Namespacerrrrrr s!r)rÚvllm.benchmarks.serverÚvllm_omni.benchmarks.serverÚ(vllm_omni.entrypoints.cli.benchmark.baserrrrrrÚs