Shuhaohust's picture
Re-upload with all required fields
6dc5b0f verified
metadata
license: mit
task_categories:
  - benchmarking
language:
  - en
tags:
  - llm
  - scheduling
  - control-plane
  - benchmark
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: workload_id
      dtype: string
    - name: request_count
      dtype: int64
    - name: rate_per_second
      dtype: float64
    - name: arrival_pattern
      dtype: string
    - name: model_distribution
      struct:
        - name: Qwen/Qwen2.5-14B-Instruct
          dtype: float64
        - name: Qwen/Qwen2.5-7B-Instruct
          dtype: float64
        - name: meta-llama/Llama-3.1-8B-Instruct
          dtype: float64
        - name: mistralai/Mistral-7B-Instruct-v0.3
          dtype: float64
    - name: priority_distribution
      struct:
        - name: HIGH
          dtype: float64
        - name: LOW
          dtype: float64
        - name: NORMAL
          dtype: float64
    - name: prompt_len_range
      list: int64
    - name: output_len_range
      list: int64
    - name: slo_deadlines
      struct:
        - name: HIGH
          dtype: int64
        - name: LOW
          dtype: int64
        - name: NORMAL
          dtype: int64
  splits:
    - name: train
      num_bytes: 1021
      num_examples: 7
  download_size: 8744
  dataset_size: 1021

SAGE Control Plane Workloads

Workload configurations for benchmarking Control Plane scheduling policies.

Usage

from datasets import load_dataset
workloads = load_dataset("intellistream/sage-control-plane-workloads")