# AIPerf 0.10.0+ Schema 2.0 — Extreme Context Test Template (10k Input / 10k Output)
# Use case: Stress-test context-window limits and sustained decode performance
# Concurrency is 1 because runtime is 200–1000 seconds per request

schema_version: "2.0"
benchmark:
  models:
    items:
      - name: "vllm/gemma-4-31b-it"
  endpoint:
    urls: ["https://ai.noris.de/v1"]
    type: "chat"
    streaming: true
  datasets:
    - type: "synthetic"
      name: "main"
      entries: 10
      isl:
        type: "fixed"
        value: 10000
      osl:
        type: "fixed"
        value: 10000
  phases:
    - type: "concurrency"
      name: "profiling"
      requests: 10

# CLI usage (per model):
#   aiperf profile --config this-file.yaml --concurrency 1 --ui none \
#       --api-key "$API_KEY" --artifact-dir results/10k/MODEL_NAME/

# Expected runtime:
#   At 10–50 tok/s output, one request = 200–1000 seconds
#   10 requests sequentially = ~30–60 minutes per model
#   Schedule with no-concurrency (--concurrency 1) to avoid timeouts