FIPO: Eliciting Deep Reasoning with Future-KL Influenced Policy Optimization
FIPO enhances reinforcement learning for language models by using discounted future-KL divergence to improve credit assignment and extend reasoning chains, achieving better mathematical problem-solving performance. (168 upvotes on HuggingFace)
Abstract
FIPO enhances reinforcement learning for language models by using discounted future-KL divergence to improve credit assignment and extend reasoning chains, achieving better mathematical problem-solving performance.
AI-generated summary
We present Future-KL Influenced Policy Optimization (FIPO), a reinforcement learning algorithm designed to overcome reasoning bottlenecks in large language models. While GRPO style training scales effectively, it typically relies on outcome-based rewards (ORM) that distribute a global advantage uniformly across every token in a trajectory. We argue that this coarse-grained credit assignment imposes a performance ceiling by failing to distinguish critical logical pivots from trivial tokens. FIPO addresses this by incorporating discounted future-KL divergence into the policy update, creating a dense advantage formulation that re-weights tokens based on their influence on subsequent trajectory behavior. Empirically, FIPO enables models to break through the length stagnation seen in standard baselines. Evaluated on Qwen2.5-32B, FIPO extends the average chain-of-thought length from roughly 4,000 to over 10,000 tokens and increases AIME 2024 Pass@1 accuracy from 50.0% to a peak of 58.0% (converging at approximately 56.0%). This outperforms both DeepSeek-R1-Zero-Math-32B (around 47.0%) and o1-mini (approximately 56.0%). Our results suggest that establishing dense advantage formulations is a vital path for evolving ORM-based algorithms to unlock the full reasoning potential of base models. We open-source our training system, built on the verl framework.
View arXiv page View PDF Project page GitHub 42 Add to collection
Get this paper in your agent:
hf papers read 2603.19835
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash
Models citing this paper 1
Datasets citing this paper 0
No dataset linking this paper
Cite arxiv.org/abs/2603.19835 in a dataset README.md to link it from this page.
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2603.19835 in a Space README.md to link it from this page.
Collections including this paper 2
Sign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
researchpaperarxiv
Exclusive | Caltech Researchers Claim Radical Compression of High-Fidelity AI Models - WSJ
<a href="https://news.google.com/rss/articles/CBMiuANBVV95cUxPc3gxMzA2bXJ4TUVDT050Z1lkNFZieHlPWmQtTmgtd0dtRUxGQU8yYkoxdlgzLXl0a3ljaVU3b1hpMzd1NWNIb3NPdjZyQ3ZpZGs1V2R2VXBncUdqZHJvNmx0NjRWNEhpSzVqY0s1VVRtWFFqMXdIenhHc0pQYXFqM25leDRzOWhzTzlPdDRDU3lWeFk3X1FUS1A4emxnN3kwaUtJWndxNHhWWVRlaDVLU052WWxNR3FXdXVVTzlLSWh2YXZNdGYxY1dBX2F3YWV0MVgyQnFUaHc3a0dlcEpfVy03dzFRbWhSbXB6QlZiZWo2TldzVnhINHhaREZrQkNhRy04b1BFSW51SWRvaEppLUZ0TTNfMmcySWc0WVBta3VnZmJsTWd5UzhWeTVDdW5iQXBBQU1zbUh2TFBSc0ozcU5RVGJNYXdEakdKaGNjVDF3ZTNfcXFXNXlaUlNucFlWNk50bGRIZU1mOV9DNXpPZGdFaVVLMllKa3dYTXpGVWNzaHA5RVBRWThkeHFmWkM4RGdPcEFuNy1seWpHZHVWQWpsWlJsMUtYMFJCOUpKWFF0anJMTDlLcw?oc=5" target="_blank">Exclusive | Caltech Researchers Claim Radical Compression of High-Fidelity AI Models</a> <font color="#6f6f6f">WSJ</font>
Inside Claude Code’s leaked source: swarms, daemons, and 44 features Anthropic kept behind flags
On Wednesday, security researcher Chaofan Shou discovered that Anthropic had shipped version 2.1.88 of Claude Code with a 59.8MB source The post Inside Claude Code’s leaked source: swarms, daemons, and 44 features Anthropic kept behind flags appeared first on The New Stack .
Washington Paid Daycares First, Checked Papers Later: $37M in Questionable Payments Found After $416M Went Unaudited
An audit of Washington state's child care subsidy system revealed $37 million in questionable payments. Explore the oversight issues and proposed reforms to boost accountability and transparency.
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Research Papers
AI Inspires New Research Topics In Materials Science - miragenews.com
<a href="https://news.google.com/rss/articles/CBMihwFBVV95cUxQRlVFdkRBaHRvYkJJdFRlMTZmajEzeFRPU0hGWWdfbi02V1FnTUdVQ2pmY2VZLUV2NlB4V3BFdEVlSVZkUlhRSTZaNWFKMmcyWXJYbnNqbUhMTmp0NnFtMEppOXlPZkJSNHJfck5VSEVYcmUtX1k2QkJlR1BvUEdTTkp3UmlYRkk?oc=5" target="_blank">AI Inspires New Research Topics In Materials Science</a> <font color="#6f6f6f">miragenews.com</font>
From brain scans to alloys: Teaching AI to make sense of complex research data - Penn State University
<a href="https://news.google.com/rss/articles/CBMiwAFBVV95cUxPZDFHdkptQ2VUM2hmWjhqQkxoRnBiTWoxMXRRR21MUG5TamdUMlFRWmhvYVNHaFVNREVKU3VmSnVOdDVZYnNLb2ppYXRVRTZmVFVMV1pLTlVhUm9ybTNZbGtvZTdIMnIyMHNpOEk5aU9TSmxxS2Y4V2MwazYwY3JlX1Axbk1nd3pfcWhFdUJaaDJWRXJaMFIyTTROcmFHeXI3ZzFudXJ2M1h6UHI1LW1Ca1dta2RkM3BiYndocGk3Yjg?oc=5" target="_blank">From brain scans to alloys: Teaching AI to make sense of complex research data</a> <font color="#6f6f6f">Penn State University</font>
Locating Risk: Task Designers and the Challenge of Risk Disclosure in RAI Content Work
arXiv:2505.24246v4 Announce Type: replace Abstract: As AI systems are increasingly tested and deployed in open-ended and high-stakes domains, crowdworkers are often tasked with responsible AI (RAI) content work. These tasks include labeling violent content, moderating disturbing text, or simulating harmful behavior for red teaming exercises to shape AI system behaviors. While prior research efforts have highlighted the risks to worker well-being associated with RAI content work, far less attention has been paid to how these risks are communicated to workers by task designers or individuals who design and post RAI tasks. Existing transparency frameworks and guidelines, such as model cards, datasheets, and crowdworksheets, focus on documenting model information and dataset collection process
Togedule: Scheduling Meetings with Large Language Models and Adaptive Representations of Group Availability
arXiv:2505.01000v5 Announce Type: replace Abstract: Scheduling is a perennial-and often challenging-problem for many groups. Existing tools are mostly static, showing an identical set of choices to everyone, regardless of the current status of attendees' inputs and preferences. In this paper, we propose Togedule, an adaptive scheduling tool that uses large language models to dynamically adjust the pool of choices and their presentation format. With the initial prototype, we conducted a formative study (N=10) and identified the potential benefits and risks of such an adaptive scheduling tool. Then, after enhancing the system, we conducted two controlled experiments, one each for attendees and organizers (total N=66). For each experiment, we compared scheduling with verbal messages, shared c
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!