Automatic Configuration of LLM Post-Training Pipelines

LLM post-training pipelines that combine supervised fine-tuning and reinforcement learning are difficult to configure under realistic compute budgets: the configuration space is high-dimensional and heterogeneous, stages are strongly coupled, and each end-to-end evaluation is expensive. We propose AutoPipe, a budget-aware two-stage framework for configuration selection in LLM post-training. Offline, AutoPipe learns a dataset-conditioned learning-to-rank surrogate from historical runs, capturing within-dataset preferences and providing transferable guidance toward promising regions of the configuration space. Online, for a new dataset, AutoPipe uses the offline guidance to steer Bayesian optimization and models dataset-specific deviations with a Gaussian-process residual surrogate. To reduce evaluation cost, each trial is early-stopped and scored by a learned predictor that maps early training signals to a low-cost proxy for final post-training performance. Experiments on biomedical reasoning tasks show that AutoPipe consistently outperforms offline-only baselines and achieves comparable performance with the strongest online HPO baselines while using less than 10\% of their computational cost.

Efficient LLM Serving for Agentic Workflows: A Data Systems Perspective

Agentic workflows are composed of sequences of interdependent Large Language Model (LLM) calls, and they have become a dominant workload in modern AI systems. These workflows exhibit extensive redundancy from overlapping prompts and intermediate results due to speculative and parallel exploration. Existing LLM serving systems, such as vLLM, focus on optimizing individual inference calls and overlook cross-call dependencies, leading to significant inefficiencies. This paper rethinks LLM and agent serving from a data systems perspective and introduces Helium, a workflow-aware serving framework that models agentic workloads as query plans and treats LLM invocations as first-class operators. Helium integrates proactive caching and cache-aware scheduling to maximize reuse across prompts, KV states, and workflows. Through these techniques, Helium bridges classic query optimization principles with LLM serving, achieving up to 1.56x speedup over state-of-the-art agent serving systems on various workloads. Our results demonstrate that end-to-end optimization across workflows is essential for scalable and efficient LLM-based agents.

DatedGPT: Preventing Lookahead Bias in Large Language Models with Time-Aware Pretraining

In financial backtesting, large language models pretrained on internet-scale data risk introducing lookahead bias that undermines their forecasting validity, as they may have already seen the true outcome during training. To address this, we present DatedGPT, a family of twelve 1.3B-parameter language models, each trained from scratch on approximately 100 billion tokens of temporally partitioned data with strict annual cutoffs spanning 2013 to 2024. We further enhance each model with instruction fine-tuning on both general-domain and finance-specific datasets curated to respect the same temporal boundaries. Perplexity-based probing confirms that each model's knowledge is effectively bounded by its data cutoff year, while evaluation on standard benchmarks shows competitive performance with existing models of similar scale. We provide an interactive web demo that allows users to query and compare responses from models across different cutoff years.

Portfolio Reinforcement Learning with Scenario-Context Rollout

Market regime shifts induce distribution shifts that can degrade the performance of portfolio rebalancing policies. We propose macro-conditioned scenario-context rollout (SCR) that generates plausible next-day multivariate return scenarios under stress events. However, doing so faces new challenges, as history will never tell what would have happened differently. As a result, incorporating scenario-based rewards from rollouts introduces a reward--transition mismatch in temporal-difference learning, destabilizing RL critic training. We analyze this inconsistency and show it leads to a mixed evaluation target. Guided by this analysis, we construct a counterfactual next state using the rollout-implied continuations and augment the critic agent's bootstrap target. Doing so stabilizes the learning and provides a viable bias-variance tradeoff. In out-of-sample evaluations across 31 distinct universes of U.S. equity and ETF portfolios, our method improves Sharpe ratio by up to 76% and reduces maximum drawdown by up to 53% compared with classic and RL-based portfolio rebalancing baselines.

Towards Autonomous Memory Agents

Recent memory agents improve LLMs by extracting experiences and conversation history into an external storage. This enables low-overhead context assembly and online memory update without expensive LLM training. However, existing solutions remain passive and reactive; memory growth is bounded by information that happens to be available, while memory agents seldom seek external inputs in uncertainties. We propose autonomous memory agents that actively acquire, validate, and curate knowledge at a minimum cost. U-Mem materializes this idea via (i) a cost-aware knowledge-extraction cascade that escalates from cheap self/teacher signals to tool-verified research and, only when needed, expert feedback, and (ii) semantic-aware Thompson sampling to balance exploration and exploitation over memories and mitigate cold-start bias. On both verifiable and non-verifiable benchmarks, U-Mem consistently beats prior memory baselines and can surpass RL-based optimization, improving HotpotQA (Qwen2.5-7B) by 14.6 points and AIME25 (Gemini-2.5-flash) by 7.33 points.

ForeAct: Steering Your VLA with Efficient Visual Foresight Planning

Vision-Language-Action (VLA) models convert high-level language instructions into concrete, executable actions, a task that is especially challenging in open-world environments. We present Visual Foresight Planning (ForeAct), a general and efficient planner that guides a VLA step-by-step using imagined future observations and subtask descriptions. With an imagined future observation, the VLA can focus on visuo-motor inference rather than high-level semantic reasoning, leading to improved accuracy and generalization. Our planner comprises a highly efficient foresight image generation module that predicts a high-quality 640$\times$480 future observation from the current visual input and language instruction within only 0.33s on an H100 GPU, together with a vision-language model that reasons over the task and produces subtask descriptions for both the generator and the VLA. Importantly, state-of-the-art VLAs can integrate our planner seamlessly by simply augmenting their visual inputs, without any architectural modification. The foresight generator is pretrained on over 1 million multi-task, cross-embodiment episodes, enabling it to learn robust embodied dynamics. We evaluate our framework on a benchmark that consists of 11 diverse, multi-step real-world tasks. It achieves an average success rate of 87.4%, demonstrating a +40.9% absolute improvement over the $π_0$ baseline (46.5%) and a +30.3% absolute improvement over $π_0$ augmented with textual subtask guidance (57.1%).

PCL-Reasoner-V1.5: Advancing Math Reasoning with Offline Reinforcement Learning

We present PCL-Reasoner-V1.5, a 32-billion-parameter large language model (LLM) for mathematical reasoning. The model is built upon Qwen2.5-32B and refined via supervised fine-tuning (SFT) followed by reinforcement learning (RL). A central innovation is our proposed offline RL method, which provides superior training stability and efficiency over standard online RL methods such as GRPO. Our model achieves state-of-the-art performance among models post-trained on Qwen2.5-32B, attaining average accuracies of 90.9% on AIME 2024 and 85.6% on AIME 2025. Our work demonstrates offline RL as a stable and efficient paradigm for advancing reasoning in LLMs. All experiments were conducted on Huawei Ascend 910C NPUs.

Scaling Test-time Inference for Visual Grounding

Visual grounding is an essential capability of Visual Language Models (VLMs) to understand the real physical world. Previous state-of-the-art grounding visual language models usually have large model sizes, making them heavy for deployment and slow for inference. However, we notice that the sizes of visual encoders are nearly the same for small and large VLMs and the major difference is the sizes of the language models. Small VLMs fall behind larger VLMs in grounding because of the difference in language understanding capability rather than visual information handling. To mitigate the gap, we introduce 'Efficient visual Grounding language Models' (EGM): a method to scale the test-time computation (#generated tokens). Scaling the test-time computation of a small model is deployment-friendly, and yields better end-to-end latency as the cost of each token is much cheaper compared to directly running a large model. On the RefCOCO benchmark, our EGM-Qwen3-VL-8B demonstrates 91.4 IoU with an average of 737ms (5.9x faster) latency while Qwen3-VL-235B demands 4,320ms to achieve 90.5 IoU. To validate our approach's generality, we further set up a new amodal grounding setting that requires the model to predict both the visible and occluded parts of the objects. Experiments show our method can consistently and significantly improve the vanilla grounding and amodal grounding capabilities of small models to be on par with or outperform the larger models, thereby improving the efficiency for visual grounding.

AscendKernelGen: A Systematic Study of LLM-Based Kernel Generation for Neural Processing Units

To meet the ever-increasing demand for computational efficiency, Neural Processing Units (NPUs) have become critical in modern AI infrastructure. However, unlocking their full potential requires developing high-performance compute kernels using vendor-specific Domain-Specific Languages (DSLs), a task that demands deep hardware expertise and is labor-intensive. While Large Language Models (LLMs) have shown promise in general code generation, they struggle with the strict constraints and scarcity of training data in the NPU domain. Our preliminary study reveals that state-of-the-art general-purpose LLMs fail to generate functional complex kernels for Ascend NPUs, yielding a near-zero success rate. To address these challenges, we propose AscendKernelGen, a generation-evaluation integrated framework for NPU kernel development. We introduce Ascend-CoT, a high-quality dataset incorporating chain-of-thought reasoning derived from real-world kernel implementations, and KernelGen-LM, a domain-adaptive model trained via supervised fine-tuning and reinforcement learning with execution feedback. Furthermore, we design NPUKernelBench, a comprehensive benchmark for assessing compilation, correctness, and performance across varying complexity levels. Experimental results demonstrate that our approach significantly bridges the gap between general LLMs and hardware-specific coding. Specifically, the compilation success rate on complex Level-2 kernels improves from 0% to 95.5% (Pass@10), while functional correctness achieves 64.3% compared to the baseline's complete failure. These results highlight the critical role of domain-specific reasoning and rigorous evaluation in automating accelerator-aware code generation.

Coding in a Bubble? Evaluating LLMs in Resolving Context Adaptation Bugs During Code Adaptation

Code adaptation is a fundamental but challenging task in software development, requiring developers to modify existing code for new contexts. A key challenge is to resolve Context Adaptation Bugs (CtxBugs), which occurs when code correct in its original context violates constraints in the target environment. Unlike isolated bugs, CtxBugs cannot be resolved through local fixes and require cross-context reasoning to identify semantic mismatches. Overlooking them may lead to critical failures in adaptation. Although Large Language Models (LLMs) show great potential in automating code-related tasks, their ability to resolve CtxBugs remains a significant and unexplored obstacle to their practical use in code adaptation. To bridge this gap, we propose CtxBugGen, a novel framework for generating CtxBugs to evaluate LLMs. Its core idea is to leverage LLMs' tendency to generate plausible but context-free code when contextual constraints are absent. The framework generates CtxBugs through a four-step process to ensure their relevance and validity: (1) Adaptation Task Selection, (2) Task-specific Perturbation,(3) LLM-based Variant Generation and (4) CtxBugs Identification. Based on the benchmark constructed by CtxBugGen, we conduct an empirical study with four state-of-the-art LLMs. Our results reveal their unsatisfactory performance in CtxBug resolution. The best performing LLM, Kimi-K2, achieves 55.93% on Pass@1 and resolves just 52.47% of CtxBugs. The presence of CtxBugs degrades LLMs' adaptation performance by up to 30%. Failure analysis indicates that LLMs often overlook CtxBugs and replicate them in their outputs. Our study highlights a critical weakness in LLMs' cross-context reasoning and emphasize the need for new methods to enhance their context awareness for reliable code adaptation.