You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
title = "Can RL-based LLM post-training on games generalize to other tasks? (GRL)"
3
+
date = 2025-08-27T12:00:00-08:00
4
+
authors = ["Game Arena Team"]
5
+
author = "Game Arena Team"
6
+
ShowReadingTime = true
7
+
draft = false
8
+
redirect_url = "https://lmgame.org/#/blog/grl"
9
+
[cover]
10
+
image = "img/lmgame/grl_system_design.png"
11
+
alt = "grl"
12
+
caption = "Post-training LLMs on games (Sokoban, Tetris) improves same-family variants (≈ +2–56%) and shows smaller gains on related tasks (Blocksworld +3–7%, WebShop ~+6% but unstable); no improvement on GSM8K. We introduce GRL, an agent-centric, multi-turn RL framework that makes LLM–environment interaction highly customizable for systematic generalization studies."
caption = "This guideline contains a detailed overview of the repository setup of Lmgame-Bench as well as how to use it for evaluation as well as the integration of new games. It covers the process of evaluation in the single/multi-agent mode as well as the various LLMs that you can use."
caption = "Pokémon is increasingly used to evaluate modern large language models, but current practices lack standardization, depend heavily on game-specific scaffolding, and are costly. We address these issues with lmgame-bench, a new framework offering standardized evaluations and initial results across diverse games."
Copy file name to clipboardExpand all lines: content/home.md
+4Lines changed: 4 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -32,6 +32,10 @@ We also develop and maintain open-source models, evaluations, and systems to dem
32
32
33
33
<!-- <div class="news-year">2025</div> -->
34
34
35
+
{{< newsEntry date="2025/06" >}}
36
+
Hao was awarded with [Google ML and Systems Junior Faculty Award](https://datascience.ucsd.edu/hdsi-assistant-professor-hao-zhang-receives-google-junior-faculty-award/)!
37
+
{{< /newsEntry >}}
38
+
35
39
{{< newsEntry date="2025/04" >}}
36
40
NVIDIA gifted us a DGX B200! See [our recent post](https://x.com/haoailab/status/1914402516420440072) here.
0 commit comments