Back to evaluations
Public evaluation
kv_batching_task
Simulates continuous request traffic to evaluate queue admission decisions and KV cache memory utilization.
Evaluation type
task based
Challenge
Continuous Batching Memory Planner with Google ADK and LlamaIndex
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
kv_batching_task
Tests batch allocation strategy under simulated peak load.
Input format
JSON containing hardware_spec (total_vram_gb, page_size_mb) and concurrent_requests array
Output format
JSON detailing scheduled_batch_size, admitted_request_ids, rejected_request_ids, and estimated_p99_ms