Back to evaluations
Public evaluation
continuous_batch_eval
Evaluates continuous batching efficiency and power limit adherence under concurrent token generation tasks.
Evaluation type
task based
Challenge
Google ADK & Upstage Continuous Batching Scheduler
Difficulty
Advanced
Rigor
Unspecified
Evaluation overview
How the linked challenge is judged: tasks, benchmarks, and criteria count.
Tasks
1
Benchmarks
0
Criteria
0
Task templates
Inputs and expected outputs.
Task 1
continuous_batch_eval
Schedules 100 incoming inference requests under a 700W node power cap.
Input format
JSON containing request_queue, power_cap_watts, max_batch_tokens
Output format
JSON containing total_throughput_tok_s, avg_ttft_ms, avg_itl_ms, power_peak_watts