---
title: "Discussion on \"Multi-Reward RL, Part 2: Benchmarking GRPO, DAPO, and CISPO on Unseen Tasks\" | Hashnode"
description: "Discussion on \"Multi-Reward RL, Part 2: Benchmarking GRPO, DAPO, and CISPO on Unseen Tasks\". Qwen3-14B DEX experiments compare GRPO, DAPO, GDPO, CISPO, ADAPO and REPO-R with and without thinking: full holdout results, reward curves, time and memory."
url: "https://hashnode.com/posts/multi-reward-rl-part-2-benchmarking-grpo-dapo-and-cispo-on-unseen-tasks/6ac5597f65f0cea9c5cc433d"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Multi-Reward RL, Part 2: Benchmarking GRPO, DAPO, and CISPO on Unseen Tasks](https://gfactor.hashnode.dev/multi-reward-rl-part-2-benchmarking-grpo-dapo-and-cispo-on-unseen-tasks?utm_source=hashnode&utm_medium=feed)

Qwen3-14B DEX experiments compare GRPO, DAPO, GDPO, CISPO, ADAPO and REPO-R with and without thinking: full holdout results, reward curves, time and memory.

Oct 6 · 14 min read · gfactor.hashnode.dev  
Read the article

-   [#reinforcement-learning](https://hashnode.com/tag/reinforcement-learning)
-   [#llm](https://hashnode.com/tag/llm)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#artificial-intelligence](https://hashnode.com/tag/artificial-intelligence)
-   [#benchmark](https://hashnode.com/tag/benchmark)

## Written by

[Aleksei Romanov](https://hashnode.com/@gfactor)

@gfactor · United States

  
[View profile](https://hashnode.com/@gfactor)

Ex-Apple GenAI engineer, building an LLM post-training platform at g factor technologies

Ex-Apple GenAI engineer, building an LLM post-training platform at g factor technologies

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
