Back to evaluations
Public evaluation
kv_capacity_eval
Tests KV-cache budget calculations and admission control decision validity under high memory stress.
Evaluation type
task based
Challenge
CrewAI & Agno Long-Context KV-Cache Capacity Planner
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_capacity_eval
Evaluates memory allocation correctness for 1M token context requests across 8x H100 GPUs.
Input format
JSON containing gpu_memory_gb, num_layers, num_heads, num_kv_heads, head_dim, active_requests
Output format
JSON containing max_concurrency, required_kv_cache_gb, admission_decision