Talos is an open-source tool that automates iterative reanalysis of genomic data to identify rare disease diagnoses. It achieved a 90% recovery rate of in-scope diagnoses with only 1.3 candidate variants per patient, and delivered 241 new diagnoses across 5,000 undiagnosed patients, with most new findings emerging within 32 days of evidence publication.
UltraQuant enables 4-bit KV caching for context-heavy agents, reducing P50 time-to-first-token by 3.47x in late rounds and boosting output throughput by 1.63x over FP8 KV baseline. It achieves this using FP8 queries, FP4 KV tensors, UE8M0 group scales, and native scaled-MFMA on AMD CDNA4 GPUs, with optimizations for decode-attention kernels and robust design choices like asymmetric K/V treatment and Walsh-Hadamard rotation.
UltraQuant introduces a 4-bit KV caching method tailored for context-heavy agent workloads. It achieves 3.47x reduction in P50 time-to-first-token in late rounds and 1.63x higher output throughput compared to FP8 KV caching, using FP8 queries, FP4 KV tensors, and native AMD CDNA4 scaled-MFMA support.
A new method called probe-and-refine tuning uses synthetic bug-fix probes to iteratively improve repository guidance files with single-shot LLM calls, without agent loops or tool use. On SWE-bench Verified, it achieves a 33.0% mean resolve rate—14.5 percentage points higher than the initial static knowledge base—showing improved coverage rather than patch precision. The method enables agents to use larger step budgets effectively, and performance remains stable across models when diagnostic output is sufficient.
Execution-state capsules enable graph-bound checkpointing and restoration of complete execution state, including KV, recurrent, and convolution states, for low-latency, small-batch on-device AI serving. On RTX 5090 and Jetson AGX Thor, capsule restore achieves byte-exact and token-identical correctness, with sub-millisecond GPU operations and TTFT speedups up to 27x at 16k tokens, demonstrating significant latency reduction in interactive AI workflows.
H-RePlan introduces a hierarchical replanning framework that separates device-local strategy recovery from global orchestrator replanning. It outperforms existing baselines by achieving higher completion and instruction adherence, with reduced token cost, through scope-aware recovery in multi-device agent systems.
Vision-Language-Action models show severe layer-wise redundancy despite large parameter counts. A training-free compression method using Centered Kernel Alignment removes twin layers, reducing model depth by up to 50% and enabling 40-50% faster training and up to 30% faster inference without performance loss, validated across simulation and real-world robotic tasks.
AutoPass uses runtime and compiler evidence to guide LLM-generated optimization decisions, outperforming expert heuristics and classical autotuning methods. It achieves geometric-mean speedups of 1.043x on x86-64 and 1.117x on ARM64 systems without prior training or fine-tuning.
CRAX introduces a high-fidelity, accelerated safety benchmark for reinforcement learning using MuJoCo XLA. It achieves up to 100x speedups over CPU-based benchmarks via vectorization and hardware acceleration, featuring six environment suites and three agent-specific tasks across three difficulty levels. Evaluation of six safe RL methods shows no single approach dominates, highlighting trade-offs between performance and safety, with curriculum learning and safety transfer improving results.
Pose6DAug enables robot data augmentation by swapping objects in successful episodes while preserving physically valid 6D pose trajectories. It operates in 3D using a mesh anchored by temporally coherent poses, ensuring multi-view consistency and physical plausibility. Fine-tuning a VLA policy on this augmented data improves novel object success rates by 16.5% over state-of-the-art baselines.
CRAX introduces a high-fidelity, fast safety benchmark for reinforcement learning using MuJoCo XLA. It achieves up to 100x speedups over CPU-based benchmarks via vectorization and hardware acceleration, featuring six environment suites and three agent-specific tasks across three difficulty levels. Evaluation of six safe RL methods shows no single approach dominates, highlighting trade-offs between performance and safety, with curriculum learning and safety transfer improving results.
JAMER introduces JamSet and JamBench, the first project-level game code dataset and benchmark on a professional game engine. Built from 8,133 verified Game Jam projects, it enables deterministic evaluation and reveals a capability cliff in AI models as project scale increases, with runtime pass rates dropping from 80.4% to 5.7%.
User as Engram proposes storing per-user facts as surgical, hash-keyed edits to a memory table, leaving reasoning in a shared adapter. This design achieves 5.6x higher indirect-reasoning accuracy and maintains base-level reasoning performance, with a memory footprint 33,000x smaller than per-user LoRA. The approach enables disjoint user edits that compose losslessly, outperforming retrieval pipelines beyond 100 facts.
Data Intelligence Agents (DIA) deploy autonomous coding agents to streamline enterprise data workflows. The Query Generator matches or exceeds top published results on seven SQL benchmarks across four dialects, showing generalization through natural-language instructions and execution-based architecture.
ScenA conditions a text-to-audio foundation model on multiple reference voices and a natural language scene prompt to generate realistic multi-speaker conversations. It addresses the 'Reference Shortcut' issue by using a high-noise-biased training schedule, ensuring speaker assignment relies on text prompts rather than acoustic similarity. Evaluated on CoVoMix2-Dialogue, Scen- A outperforms existing systems in speaker-binding and produces rich, naturalistic audio with overlapping speech and ambient noise.
Rubric-Conditioned Self-Distillation introduces a framework that uses structured rubrics to provide fine-grained, token-level feedback during self-distillation of reasoning language models. By conditioning teacher models on rubric-level criteria, it enables more precise credit assignment than scalar rewards, outperforming GRPO and OPSD by 1.0 and 0.9 points on average across science reasoning benchmarks.
Rubric-Conditioned Self-Distillation introduces a framework that uses structured rubrics to provide fine-grained, token-level feedback during self-distillation of reasoning language models. By conditioning teacher models on rubric-level criteria, it enables more precise credit assignment than scalar rewards, outperforming GRPO and OPSD by 1.0 and 0.9 points on average across science reasoning benchmarks.
EfficientRollout introduces a self-speculative decoding framework that reduces rollout and end-to-end latency by up to 19.6% and 12.7% respectively, without compromising final model quality. It uses a quantized drafter derived from the target model and integrates a system-aware toggle policy to avoid compute-bound regimes, enabling effective speculation during evolving policy generations.
WorldLines introduces a project-driven benchmark for long-horizon embodied household assistance, capturing extended household traces with dialogues, actions, and state changes. It enables evidence-linked samples for Memory QA and Embodied Task Planning, and proposes ObsMem, an observer-grounded memory framework that supports visibility-aware memories and state-aware decisions. Experiments highlight challenges in partial observability and memory translation, with ObsMem providing a stronger reference architecture for such settings.
Spotlight enables DiT RL post-training by leveraging idle spot GPUs, reducing costs by 1.4-6.4× while achieving superior image quality. It uses stale model weights in exploration and reconfigures sequence parallelism in real time, allowing efficient GPU utilization without breaking training pipelines.