def get_trtllm_pipeline()

in scripts/benchmark_pipelines.py [0:0]


def get_trtllm_pipeline(args: Namespace):
    return pipeline(
        model=args.model,
        use_fp8=args.use_fp8,
        use_cuda_graph=args.use_cuda_graph,
        max_batch_size=args.batch_size,
        max_prompt_length=args.prompt_length,
        max_new_tokens=args.max_new_tokens,
        tp=args.tp,
        pp=args.pp,
        gpus_per_node=args.gpus_per_node,
        world_size=args.world_size,
        dtype=args.dtype,
    )