Skip to content
Papers.

Stabilizing Reinforcement Learning with LLMs: Formulation and Practices research paper by Alibaba (Qwen), 2025

Alibaba (Qwen) · Dec 1, 2025 · Inference and efficiency · 40 citations · 109 upvotes · unverified

Read on arXiv

What it shows

The paper provides a theoretical foundation for optimizing sequence-level rewards in reinforcement learning using token-level objectives, highlighting the importance of techniques like importance sampling correction, clipping, and Routing Replay for stabilizing training, especially with large language models.

UnverifiedHugging Face's summary; not yet checked by hand.

More from Alibaba (Qwen)

All 61
PaperCitations
HappyWorld-BenchEvaluating world models requires assessing both the quality of the worlds they generate and their consistency and responsiveness under exploration, interaction, and modification.Agents and evaluation · Sep 2026 · Unverified0
One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering AgentsRepository-level software engineering (SWE) comprises heterogeneous task categories, whose progress under pooled agentic reinforcement learning can be uneven: gains in some categories coincide with regressions in...Training and scaling · Sep 2026 · Unverified0
OmniVChat: Synthesizing, Benchmarking, and Training for Native Audio-Visual DialogueWe define OmniVChat (Omni Video Chat) as the task of native audio-visual dialogue between a user and an omni model.Multimodal and robotics · Sep 2026 · Unverified0
RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use AgentsComputer-use agents (CUAs) have advanced along two separate lines: graphical interaction and software development through code and the command line.Agents and evaluation · Sep 2026 · Unverified0
CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker DistillationCORE distills compositional ranking judgments from a cross-attentive reranker into an embedding model via synthesized multi-level candidates and a Rank-KL objective, improving compositional retrieval without degrading standard performance.Reasoning · Sep 2026 · Unverified0
Terminal-Universe: Turning Agent Trajectories into Scalable Terminal EnvironmentsTerminal-Universe reconstructs executable workspaces from agent trajectories to synthesize diverse training tasks and improves post-training performance through supervised fine-tuning.Agents and evaluation · Sep 2026 · Unverified1
Qwen-Drive-1.0: An Initial Step towards a Vision-Language Foundation Model for Autonomous DrivingQwen-Drive-1.0 is a vision-language foundation model for autonomous driving that unifies 3D perception, visual question answering, and motion planning via shared representations and staged training.Multimodal and robotics · Aug 2026 · Unverified3
On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training StabilityQwen3.8-Flash-Next: a 125B mixture-of-experts model with 6B active that nearly matches its 397B predecessor at 1/9 the training compute.Architectures · Aug 20266

Other inference and efficiency papers

Topic
PaperCitations
Efficient Memory Management for Large Language Model Serving with PagedAttentionvLLM manages the KV cache like pages of virtual memory, serving models with 2 to 4 times the throughput.UC Berkeley · Sep 20238,481
FlashAttention: Fast and Memory-Efficient Exact Attention with IO-AwarenessFlashAttention computes exact attention with far fewer GPU memory reads and writes, making long sequences faster.Stanford University · May 20225,334
Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse AttentionNSA, a trainable sparse attention mechanism, enhances long-context modeling efficiency without sacrificing performance, achieving improvements in speed and accuracy over full attention models.DeepSeek · Feb 2025 · Unverified507
SmolVLM: Redefining small and efficient multimodal modelsSmolVLM, a series of compact multimodal models, achieves high performance with minimal GPU memory usage, making efficient deployment on mobile and edge devices possible.Hugging Face · Apr 2025 · Unverified294
Inference-Time Scaling for Generalist Reward ModelingSelf-Principled Critique Tuning enhances pointwise generative reward modeling for large language models, improving scalability and quality compared to existing methods.DeepSeek · Apr 2025 · Unverified249
DeepSeek-OCR: Contexts Optical CompressionDeepSeek-OCR uses optical 2D mapping to compress long contexts, achieving high OCR precision with reduced vision tokens and demonstrating practical value in document processing.DeepSeek · Oct 2025 · Unverified189
About this paper
Authors
Chujie Zheng, Kai Dang, Bowen Yu and 7 more
arXiv
2512.01374 · PDF
Venue
arXiv.org
Citations
40, 4 influential · Semantic Scholar
Upvotes
109 · Hugging Face
Lab
Alibaba (Qwen) · on Companies · on Quarterly

Changes

What changed
Influential citationsfirst count: 4Sep 25, 2026
Citationsfirst count: 40Sep 25, 2026
New paperFound by the weekly scan, unverifiedSep 25, 2026

Sources: each lab's own papers and arXiv, with citation and upvote counts from Semantic Scholar and Hugging Face. One-line summaries are for orientation, not a substitute for the paper. Logos via logo.dev; trademarks belong to their owners.

New papers by email

Monday afternoons, only in weeks with new papers from the labs.

Double opt-in. Unsubscribe any time.