[EXAMPLE] poker inference script with two models

[burtenshaw]

I've been trying out the library and implemented this simple script to test models on the kuhn_poker task in open spiel.

It tests two models on the task, using the openai inference client, so it's provider agnostic.I used it with HF inference providers, but you could vLLM, openai, etc.

The example script follows the style of the others and prints out a report:

============================================================
COMPARATIVE RESULTS
============================================================

Rank   Model                                    Avg $/game   Win Rate   Record         
------ ---------------------------------------- ------------ ---------- ---------------
🥇 1   DeepSeek-V3.1-Terminus                   +0.020         55.0%     55W-45L-0P     
🥈 2   Qwen2.5-72B-Instruct                     -0.200         40.0%     40W-60L-0P     

🏆 Best performer: DeepSeek-V3.1-Terminus
   Chips/game: +0.020
   Win rate: 55.0% (55/100)

📊 Strategy Analysis:
   DeepSeek-V3.1-Terminus           5.0% PASS /  95.0% BET
   Qwen2.5-72B-Instruct            50.5% PASS /  49.5% BET

============================================================

[pankit-eng]

do you want to use from_docker_image("ghcr...") just so the person playing with this example does not have to bring this up and this example works out of the box?

[burtenshaw]

Makes sense.

[burtenshaw]

Made the change, but I couldn't test on ghcr.io/meta-pytorch/openspiel-env:latest because it's private. I tested on a local build of openspiel-env though. Tested now.

I also opened this from the process

[init27]

@burtenshaw I just came to say you're a legend, thanks for so many fast contributions!!!!

[burtenshaw]

@burtenshaw I just came to say you're a legend, thanks for so many fast contributions!!!!

No worries @init27 . I'm having so much fun with openenv.