ReLibra uses pre-known token-to-expert routing from RL rollouts to perform inter-batch expert reordering and intra-batch replication, delivering up to 1.6x higher throughput than Megatron-LM and 1.2x over oracle-equipped EPLB while staying within 6-10% of an ideal balanced baseline.
hub Mixed citations
Generalizing Verifiable Instruction Following
Mixed citation behavior. Most common role is background (44%).
28
Pith papers citing it
Background
44% of classified citations
abstract
A crucial factor for successful human and AI interaction is the ability of language models or chatbots to follow human instructions precisely. A common feature of instructions are output constraints like ``only answer with yes or no" or ``mention the word `abrakadabra' at least 3 times" that the user adds to craft a more useful answer. Even today's strongest models struggle with fulfilling such constraints. We find that most models strongly overfit on a small set of verifiable constraints from the benchmarks that test these abilities, a skill called precise instruction following, and are not able to generalize well to unseen output constraints. We introduce a new benchmark, IFBench, to evaluate precise instruction following generalization on 58 new, diverse, and challenging verifiable out-of-domain constraints. In addition, we perform an extensive analysis of how and on what data models can be trained to improve precise instruction following generalization. Specifically, we carefully design constraint verification modules and show that reinforcement learning with verifiable rewards (RLVR) significantly improves instruction following. In addition to IFBench, we release 29 additional new hand-annotated training constraints and verification functions, RLVR training prompts, and code.
hub tools
citation-role summary
background 5
dataset 3
method 1
citation-polarity summary
representative citing papers
User-turn generation reveals that LLMs' interaction awareness is largely decoupled from task accuracy, remaining near zero in deterministic settings even as accuracy scales to 96.8% on GSM8K.
IFMTBench is a new benchmark for multilingual translation instruction following that tests models on single and multi-constraint scenarios using deterministic checkers and LLM judges.
Game-Time Benchmark shows spoken language models handle basic tasks but degrade sharply under temporal constraints like tempo adherence and synchronized responses.
Think-with-Rubrics has LLMs generate rubrics internally before responding, outperforming external rubric-as-reward baselines by 3.87 points on average across benchmarks.
Star Elastic trains N nested submodels in a single post-training job on a parent reasoning LLM, supporting elastic budget control that matches or exceeds independent baselines while cutting training compute by up to 360x.
SOB benchmark shows LLMs achieve near-perfect schema compliance but value accuracy of only 83% on text, 67% on images, and 24% on audio.
CompliBench uses simulation and adversarial flaw injection to create labeled dialogue data showing that top proprietary LLMs perform poorly at spotting guideline violations while fine-tuned smaller models outperform them and generalize to new domains.
ManyIH and ManyIH-Bench address instruction conflicts in LLM agents with up to 12 privilege levels across 853 tasks, revealing frontier models achieve only ~40% accuracy.
This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.
IFCodeEvolve synthesizes coding data via actor-schema co-evolution with MCTS, boosting a 32B model's performance to match proprietary SOTA on instruction following.
BINEVAL turns evaluation criteria into binary questions answered by LLMs to yield transparent multi-dimensional scores that match human judgments on standard benchmarks and support iterative prompt optimization.
RLBFF extracts binary principles from human feedback to train reward models that outperform Bradley-Terry models on RM-Bench and JudgeBench and enable customizable inference-time focus for LLM alignment.
FlashEvolve accelerates LLM agent self-evolution via asynchronous stage orchestration and inspectable language-space staleness handling, reporting 3.5-4.9x proposal throughput gains over synchronous baselines on GEPA workloads.
SEIF creates a self-reinforcing loop in which an LLM alternately generates increasingly difficult instructions and learns to follow them better using reinforcement learning signals from its own judgments.
AutoPyVerifier learns compact sets of executable Python verifiers from labeled LLM outputs via LLM synthesis and DAG search, improving objective prediction by up to 55 F1 points and downstream LLM accuracy by up to 17 points.
GroupDPO decouples group-wise preference optimization during backpropagation to cut peak memory while keeping the same gradients, allowing larger groups and consistent gains over single-pair DPO plus an NLL term on positives.
RTT bridges response-level rubrics to token-level rewards via a relevance discriminator and intra-sample group normalization, yielding higher instruction and rubric accuracy than baselines.
CARE-RL combines PA-GRM for task-adaptive rewards on open-ended tasks and DACSP for modulating RL updates using historical capability directions, reporting higher total average scores than baselines on Qwen models.
NVIDIA releases the Nemotron 3 model family with hybrid Mamba-Transformer architecture, LatentMoE, NVFP4 training, MTP layers, and multi-environment RL post-training for reasoning and agentic tasks.
Qwen3.5-Omni scales an omnimodal model to hundreds of billions of parameters with 256k context, introduces ARIA for stable speech synthesis, and reports SOTA performance on 215 audio-visual benchmarks while adding multilingual and audio-visual coding capabilities.