Skip to content

Benchmark: Add Long Prompt and Cache Simulation Fixtures #8

Description

@ron2111

Summary

Add benchmark/test fixtures for long repeated prompts and dynamic-prefix cache breakers.

Cases

  • stable 10k-token system prefix plus dynamic user suffix
  • timestamp inserted before stable prefix
  • changing tool schema order
  • model switch between turns
  • long RAG context before/after dynamic fields

Purpose

These fixtures will support future cache-simulation and recommendation work.

Acceptance Criteria

  • Fixtures are deterministic and small enough for the repo.
  • Benchmarks avoid network calls and provider credentials.
  • Results help validate future cache-sim logic.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions