GoLongRL: Capability-Oriented Long Context RL with Multitask Alignment

pbd1 pts0 comments

GitHub - xiaoxuanNLP/GoLongRL: GoLongRL: Capability-Oriented Long Context Reinforcement Learning with Multitask Alignment 路 GitHub

/" data-turbo-transient="true" />

Skip to content

Search or jump to...

Search code, repositories, users, issues, pull requests...

-->

Search

Clear

Search syntax tips

Provide feedback

--><br>We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

-->

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

Sign in

/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up

Appearance settings

Resetting focus

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

xiaoxuanNLP

GoLongRL

Public

Notifications<br>You must be signed in to change notification settings

Fork

Star<br>47

main

BranchesTags

Go to file

CodeOpen more actions menu

Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit

History<br>17 Commits<br>17 Commits

Eval/QwenLong-Benchmarks

Eval/QwenLong-Benchmarks

Train/verl

Train/verl

assets

assets

README.md

README.md

View all files

Repository files navigation

馃幆 GoLongRL: Capability-Oriented Long Context Reinforcement Learning with Multitask Alignment

馃摚 Latest News

[May 19, 2026] 馃摙 GoLongRL is available on arXiv and HuggingFace Daily.

馃搶 Overview

We present GoLongRL , a fully open-source, capability-oriented post-training recipe for long-context reinforcement learning with verifiable rewards (RLVR).

Existing long-context RL methods tend to focus data construction on retrieval-path complexity鈥攎ulti-hop chains, UUID tracking, chunk-based QA鈥攚hile collapsing diverse task objectives into a single binary reward. This leaves critical capabilities such as summarization, ranking, aggregation, and structured reasoning without direct training signal. GoLongRL addresses these limitations through two contributions.

GoLongRL addresses these limitations through two key contributions. (1) Capability-oriented data construction , a 23K-sample RLVR dataset spanning 9 task types with heterogeneous reward functions, and (2) TMN-Reweight , a multitask optimization method that combines task-level mean normalization for cross-task reward scale alignment with difficulty-adaptive reweighting for more reliable advantage estimation.

Overall performance comparison on long-context benchmarks (DocMath, LongBench-V2, Frames, MRCR, CorpusQA, LBV1-QA).

GoLongRL-30B-A3B achieves long-context performance comparable to DeepSeek-R1-0528 and Qwen3-235B-A22B-Thinking-2507, while using a significantly smaller activated parameter budget.

Model<br>Avg.<br>DocMath<br>LBV2<br>Frames<br>MRCR<br>CorpusQA<br>LBV1-QA

Qwen3-30B-A3B-Thinking-2507<br>60.1<br>63.3<br>48.7<br>70.2<br>41.6<br>70.5<br>66.5

DeepSeek-R1-0528<br>68.7<br>63.4<br>59.5<br>76.9<br>64.9<br>77.5<br>69.9

Qwen3-235B-A22B-Thinking<br>68.5<br>65.8<br>57.5<br>75.1<br>66.2<br>75.3<br>70.9

Gemini-2.5-Flash-Thinking<br>68.7<br>64.8<br>56.8<br>65.8<br>78.8<br>79.4<br>66.9

QwenLong-L1.5 (w. GRPO)<br>67.2<br>65.1<br>55.3<br>71.4<br>66.9<br>76.9<br>67.9

GoLongRL-30B-A3B (Ours)<br>69.8<br>65.3<br>55.1<br>74.5<br>81.6<br>73.6<br>68.7

Our framework combines the following.

Capability-Oriented Dataset (23K samples, 9 task types). Guided by a taxonomy of long-context capabilities, the dataset covers precise retrieval, comprehension, exhaustive retrieval, numerical reasoning, structured extraction, structured matching, graded ranking, sequence ordering, and summarization. Each task is paired with its natural evaluation metric (EM, Accuracy, F1, math_verify, IoU, SubEM, NDCG, Pairwise, ROUGE-L) as the reward function, rather than being collapsed into a single indicator.

TMN-Reweight. When training on heterogeneous reward types, per-prompt normalization in standard GRPO can mix up cross-task scale differences with prompt difficulty. TMN-Reweight is a simple modification that normalizes advantages at the task level instead of the prompt level, and adds a difficulty-adaptive weight to reduce noise from very easy or very hard prompts. It provides a modest but consistent improvement over vanilla GRPO in our ablations (+0.8 avg. at 4B scale), with gains mainly on aggregation-intensive benchmarks like CorpusQA.

Full Open Release. We publicly release the complete dataset and all training code.

Key Results

Under the same vanilla GRPO setup, our dataset alone outperforms the closed-source QwenLong-L1.5 dataset at both 4B and 30B scales (+6.1 avg. at 4B, +2.6 avg. at 30B).

TMN-Reweight further improves average performance to 63.0 at 4B scale, surpassing QwenLong-L1.5 with its specialized AEPO algorithm (59.4).

General capabilities (MMLU-Pro, AIME24/25, GPQA) are preserved or improved, with substantial gains on dialogue memory (LongMemEval +13.6) and agentic memory...

golongrl long context task capability oriented

Related Articles