mirts-bench 0.3.0

MiniRTS LLM Benchmark

MiniRTS LLM Benchmark

Deterministic 11x11 grid RTS benchmark for LLMs.
No OCR, no UI automation, fast and reproducible.
Strict 30 action cap.

Install

pip install mirts-bench

Run

polybench run --difficulty hard --opponents 7 --games 1

LLM command (stdin prompt → stdout JSON):

polybench run --difficulty hard --opponents 7 --games 1 --llm-cmd "python examples\\echo_llm.py"

OpenAI-compatible HTTP endpoint:

polybench run --difficulty hard --opponents 7 --games 1 --llm-host http://localhost:8000 --llm-model your-model --llm-api-key YOUR_KEY

Env alternative:

$env:POLYBENCH_LLM_HOST = "http://localhost:8000"
$env:POLYBENCH_LLM_MODEL = "your-model"
$env:POLYBENCH_LLM_API_KEY = "YOUR_KEY"
polybench run --difficulty hard --opponents 7 --games 1

Kaggle (local, no ngrok)

Run inside a Kaggle notebook with kaggle-benchmarks available.

!pip -q install kaggle-benchmarks mirts-bench

!polybench run --difficulty hard --opponents 7 --games 1 --llm-provider kaggle --llm-model google/gemini-2.5-flash

If --llm-model is omitted, the default Kaggle LLM is used.

Python API

import polybench

cfg = polybench.RunConfig(
    difficulty="easy",
    opponents=1,
    games=1,
    llm_host="http://localhost:8000",
    llm_model="your-model",
    llm_api_key="YOUR_KEY",
)
polybench.run_benchmark(cfg)

How it works

• Generates a deterministic 11x11 grid with cities, resources, and blockers.
• Builds a prompt from the current state.
• LLM returns a JSON action.
• The engine applies the action and advances the turn.
• Stops after 30 actions and scores the result.

Action schema

Allowed action types:

• end_turn
• move
• attack
• train
• build
• research

Unit and city identifiers are grid coordinates:

{ "type": "move", "unit_id": { "x": 3, "y": 5 }, "to": { "x": 4, "y": 5 } }

Output

Each run writes:

• runs/<timestamp>_<difficulty>_<opponents>/game_###/turn_###_prompt.txt
• runs/<timestamp>_<difficulty>_<opponents>/game_###/turn_###_response.txt
• runs/<timestamp>_<difficulty>_<opponents>/game_###/turn_###_action.json
• runs/<timestamp>_<difficulty>_<opponents>/summary.json

Options

polybench run supports:

• --difficulty easy | normal | hard | crazy
• --opponents 1 | 7 | 15
• --games (default 1)
• --llm-cmd external command that reads prompt on stdin and returns JSON on stdout
• --llm-provider openai | kaggle
• --llm-host HTTP base URL (OpenAI-compatible /v1/chat/completions)
• --llm-model model name for HTTP LLM
• --llm-api-key API key for HTTP LLM (or set POLYBENCH_LLM_API_KEY)
• --k-factor ELO K (default 32)
• --opponent-elo (default 1000)
• --start-elo (default 1000)