arXiv subjects

Minjia Zhang

Publications and source records attributed to Minjia Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

AutoSP: Unlocking Long-Context LLM Training Via Compiler-Based Sequence Parallelism

Large-language-models (LLMs) demonstrate enormous utility in long-context tasks which require processing prompts that consist of tens to hundreds of thousands of tokens. However, existing LLM training libraries do not provide easy to use abstractions to optimize for long-context training, instead focusing on optimizations for models with large parameter counts through ZeRO-3/FSDP, Tensor and Pipeline parallelism. This forces users to rewrite LLM training libraries to incorporate compositions of various complex long-context optimizations, such as sequence-parallelism, to training pipelines; a process that requires in-depth expertise, reducing developer productivity. To tackle these challenges, we introduce AutoSP: the first automated solution to automatically optimize LLM training for longer-contexts. AutoSP compiles models and applies a targeted set of optimizations: automated sequence parallelism, and long-context aware activation-checkpointing, to drastically enhance LLM trainability at negligible cost to throughput. Our evaluation demonstrates AutoSP's capability on both NVIDIA and AMD hardware, increasing training contexts by upto 2.7$\times$ and 2.5$\times$ respectively over competitive hand-written baseline at negligible cost to runtime performance.

cs.LG↗

arXiv2026

Hidden States as Early Signals: Step-level Trace Evaluation and Pruning for Efficient Test-Time Scaling

Large Language Models (LLMs) can enhance reasoning capabilities through test-time scaling by generating multiple traces. However, the combination of lengthy reasoning traces with multiple sampling introduces substantial computation and high end-to-end latency. Prior work on accelerating this process has relied on similarity-based or confidence-based pruning, but these signals do not reliably indicate trace quality. To address these limitations, we propose STEP: Step-level Trace Evaluation and Pruning, a novel pruning framework that evaluates reasoning steps using hidden states and dynamically prunes unpromising traces during generation. We train a lightweight step scorer to estimate trace quality, and design a GPU memory-aware pruning strategy that triggers pruning as the GPU memory is saturated by KV cache to reduce end-to-end latency. Experiments across challenging reasoning benchmarks demonstrate that STEP reduces end-to-end inference latency by 45%-70% on average compared to self-consistency while also improving reasoning accuracy. Our code is released at: https://github.com/Supercomputing-System-AI-Lab/STEP

cs.LG↗

arXiv2026

A Goal Without a Plan Is Just a Wish: Efficient and Effective Global Planner Training for Long-Horizon Agent Tasks

Agents based on large language models (LLMs) struggle with brainless trial-and-error and generating hallucinatory actions due to a lack of global planning in long-horizon tasks. In this paper, we introduce a plan-and-execute framework and propose EAGLET, an efficient and effective planner training method to enhance the executor agent's planning abilities without human effort. Specifically, we train a plug-and-play global planner through a two-step process: we first synthesize high-quality plans from an advanced LLM using our proposed homologous consensus filtering strategy, and apply fine-tuning as a cold start. Moreover, we further improve the planner with a rule-based reinforcement learning stage using a novel executor capability gain reward, ensuring it can handle task instructions of varying difficulty. Experiments on three long-horizon agent tasks show that executor agents equipped with our planner outperform existing methods, achieving new state-of-the-art performance. Meanwhile, EAGLET reduces training costs by 8x compared to RL-based baselines, and it does not require manual effort or extra training data, offering an efficient and effective solution.

cs.CL↗

arXiv2026

FaithLens: Detecting and Explaining Faithfulness Hallucination

Recognizing whether outputs from large language models (LLMs) contain faithfulness hallucination is crucial for real-world applications, e.g., retrieval-augmented generation and summarization. In this paper, we introduce FaithLens, a cost-efficient and effective faithfulness hallucination detection model that can jointly provide binary predictions and corresponding explanations to improve trustworthiness. To achieve this, we first synthesize training data with explanations via advanced LLMs and apply a well-defined data filtering strategy to ensure label correctness, explanation quality, and data diversity. Subsequently, we fine-tune the model on these well-curated training data as a cold start and further optimize it with rule-based reinforcement learning, using rewards for both prediction correctness and explanation quality. Results on 12 diverse tasks show that the 8B-parameter FaithLens outperforms advanced models such as GPT-5.2 and o3. Also, FaithLens can produce high-quality explanations, delivering a distinctive balance of trustworthiness, efficiency, and effectiveness.

cs.CL↗

arXiv2026

Slow-Fast Policy Optimization: Reposition-Before-Update for LLM Reasoning

Reinforcement learning (RL) has become central to enhancing reasoning in large language models (LLMs). Yet on-policy algorithms such as Group Relative Policy Optimization (GRPO) often suffer in early training: noisy gradients from low-quality rollouts lead to unstable updates and inefficient exploration. We introduce Slow-Fast Policy Optimization (SFPO), a simple yet efficient framework to address the above limitations via decomposing each step into three stages: a short fast trajectory of inner steps on the same batch, a reposition mechanism to control off-policy drift, and a final slow correction. This reposition-before-update design preserves the objective and rollout process unchanged, making SFPO plug-compatible with existing policy-gradient pipelines. Extensive experiments demonstrate that SFPO consistently improves stability, reduces number of rollouts, and accelerates convergence of reasoning RL training. Specifically, it outperforms GRPO by up to 2.80 points in average on math reasoning benchmarks. It also achieves up to 4.93\texttimes{} fewer rollouts and an up to 4.19\texttimes{} reduction in wall-clock time to match GRPO's best accuracy. Project website is available at https://slow-fast-po.github.io/.

cs.LG↗

arXiv2026

VTool-R1: VLMs Learn to Think with Images via Reinforcement Learning on Multimodal Tool Use

Reinforcement Learning Finetuning (RFT) has significantly advanced the reasoning capabilities of large language models (LLMs) by enabling long chains of thought, self-correction, and effective tool use. While recent works attempt to extend RFT to vision-language models (VLMs), these efforts largely produce text-only reasoning conditioned on static image inputs, falling short of true multimodal reasoning in the response. In contrast, test-time methods like Visual Sketchpad incorporate visual steps but lack training mechanisms. We introduce VTool-R1, the first framework that trains VLMs to generate multimodal chains of thought by interleaving text and intermediate visual reasoning steps. VTool-R1 integrates Python-based visual editing tools into the RFT process, enabling VLMs to learn when and how to generate visual reasoning steps that benefit final reasoning. Trained with outcome-based rewards tied to task accuracy, our approach elicits strategic visual tool use for reasoning without relying on process-based supervision. Experiments on structured visual question answering over charts and tables show that VTool-R1 enhances reasoning performance by teaching VLMs to "think with images" and generate multimodal chain of thoughts with tools. To support future research in multi-turn multi-modal reasoning, we open-source our code at https://github.com/VTOOL-R1/vtool-r1

cs.LG↗

arXiv2026

Aha Moment Revisited: Are VLMs Truly Capable of Self Verification in Inference-time Scaling?

Inference time techniques such as decoding time scaling and self refinement have been shown to substantially improve mathematical reasoning in large language models (LLMs), largely attributed to emergent self correction and self verification behaviors often elicited through reinforcement learning (RL). In this work, we ask whether the same recipe transfers to vision language models (VLMs), especially RL finetuned variants that claim strong visual mathematical reasoning. Through extensive evaluation, we reach three main findings that differ markedly from text only models. First, generation time capability matters more than verification and refinement: simple majority voting consistently and substantially outperforms verification centric strategies such as best of N with self verification. Second, behaviors often associated with RL tuned models at inference time, such as the 'Aha moment,' do not yield reliable reasoning performance improvements. Third, visual information is not effectively integrated into the model's self verification process. Overall, our analysis highlights a key limitation: current RL trained VLMs derive limited benefit from self verification in the visual modality, which constrains the effectiveness of inference time scaling for visual mathematical reasoning.

cs.LG↗

arXiv2025

Teaching Large Language Models to Maintain Contextual Faithfulness via Synthetic Tasks and Reinforcement Learning

Teaching large language models (LLMs) to be faithful in the provided context is crucial for building reliable information-seeking systems. Therefore, we propose a systematic framework, CANOE, to reduce faithfulness hallucinations of LLMs across different downstream tasks without human annotations. Specifically, we first synthesize short-form question-answering (QA) data with four diverse tasks to construct high-quality and easily verifiable training data without human annotation. Also, we propose Dual-GRPO, a rule-based reinforcement learning method that includes three tailored rule-based rewards derived from synthesized short-form QA data, while simultaneously optimizing both short-form and long-form response generation. Notably, Dual-GRPO eliminates the need to manually label preference data to train reward models and avoids over-optimizing short-form generation when relying only on the synthesized short-form QA data. Experimental results show that CANOE greatly improves the faithfulness of LLMs across 11 different tasks, even outperforming the most advanced LLMs, e.g., GPT-4o and OpenAI o1.

cs.CL↗

arXiv2025

SuperOffload: Unleashing the Power of Large-Scale LLM Training on Superchips

The emergence of Superchips represents a significant advancement in next-generation AI hardware. These Superchips employ a tightly coupled heterogeneous architecture that integrates GPU and CPU on the same package, which offers unprecedented computational power. However, there has been scant research investigating how LLM training benefits from this new architecture. In this work, for the first time, we study LLM training solutions based on offloading for Superchips. We observe important differences between Superchips and traditional loosely-coupled GPU-CPU architecture, which necessitate revisiting prevailing assumptions about offloading. Based on that, we present SuperOffload, a Superchip-centric offloading system that simultaneously uses Hopper GPU, Grace CPU, and NVLink-C2C interconnect more efficiently. SuperOffload accomplishes this via a combination of techniques, such as adaptive weight offloading, bucketization repartitioning, Superchip-aware casting, speculative execution, and a highly optimized Adam optimizer for Grace CPUs. Our evaluation of SuperOffload on NVIDIA GH200 demonstrates up to 2.5x throughput improvement compared to state-of-the-art offloading-based systems, enabling training of up to 25B model on a single Superchip while achieving high training throughput. We also extend SuperOffload with ZeRO-style data parallelism and DeepSpeed-Ulysses sequence parallelism, enabling training of 13B model with sequence lengths up to 1 million tokens on 8 GH200 while achieving 55% MFU.

cs.LG↗

arXiv2025

Cache-of-Thought: Master-Apprentice Framework for Cost-Effective Vision Language Model Reasoning

Vision Language Models (VLMs) have achieved remarkable success in a wide range of vision applications of increasing complexity and scales, yet choosing the right VLM model size involves a trade-off between response quality and cost. While smaller VLMs are cheaper to run, they typically produce responses only marginally better than random guessing on benchmarks such as MMMU. In this paper, we propose Cache of Thought (CoT), a master apprentice framework for collaborative inference between large and small VLMs. CoT manages high quality query results from large VLMs (master) in a cache, which are then selected via a novel multi modal retrieval and in-context learning to aid the performance of small VLMs (apprentice). We extensively evaluate CoT on various widely recognized and challenging general reasoning benchmarks, and show that CoT increases overall reasoning performance by up to 7.7% under the same budget, and specifically boosts the performance of apprentice VLMs by up to 36.6%. Our code is available at https://github.com/UIUC-MONET/Cache-of-Thoughts

cs.LG↗

arXiv2025

X-MoE: Enabling Scalable Training for Emerging Mixture-of-Experts Architectures on HPC Platforms

Emerging expert-specialized Mixture-of-Experts (MoE) architectures, such as DeepSeek-MoE, deliver strong model quality through fine-grained expert segmentation and large top-k routing. However, their scalability is limited by substantial activation memory overhead and costly all-to-all communication. Furthermore, current MoE training systems - primarily optimized for NVIDIA GPUs - perform suboptimally on non-NVIDIA platforms, leaving significant computational potential untapped. In this work, we present X-MoE, a novel MoE training system designed to deliver scalable training performance for next-generation MoE architectures. X-MoE achieves this via several novel techniques, including efficient padding-free MoE training with cross-platform kernels, redundancy-bypassing dispatch, and hybrid parallelism with sequence-sharded MoE blocks. Our evaluation on the Frontier supercomputer, powered by AMD MI250X GPUs, shows that X-MoE scales DeepSeek-style MoEs up to 545 billion parameters across 1024 GPUs - 10x larger than the largest trainable model with existing methods under the same hardware budget, while maintaining high training throughput. The source code of X-MoE is available at https://github.com/Supercomputing-System-AI-Lab/X-MoE.

cs.LG↗

arXiv2025

InstantEdit: Text-Guided Few-Step Image Editing with Piecewise Rectified Flow

We propose a fast text-guided image editing method called InstantEdit based on the RectifiedFlow framework, which is structured as a few-step editing process that preserves critical content while following closely to textual instructions. Our approach leverages the straight sampling trajectories of RectifiedFlow by introducing a specialized inversion strategy called PerRFI. To maintain consistent while editable results for RectifiedFlow model, we further propose a novel regeneration method, Inversion Latent Injection, which effectively reuses latent information obtained during inversion to facilitate more coherent and detailed regeneration. Additionally, we propose a Disentangled Prompt Guidance technique to balance editability with detail preservation, and integrate a Canny-conditioned ControlNet to incorporate structural cues and suppress artifacts. Evaluation on the PIE image editing dataset demonstrates that InstantEdit is not only fast but also achieves better qualitative and quantitative results compared to state-of-the-art few-step editing methods.

cs.CV↗

arXiv2025

FlexGaussian: Flexible and Cost-Effective Training-Free Compression for 3D Gaussian Splatting

3D Gaussian splatting has become a prominent technique for representing and rendering complex 3D scenes, due to its high fidelity and speed advantages. However, the growing demand for large-scale models calls for effective compression to reduce memory and computation costs, especially on mobile and edge devices with limited resources. Existing compression methods effectively reduce 3D Gaussian parameters but often require extensive retraining or fine-tuning, lacking flexibility under varying compression constraints. In this paper, we introduce FlexGaussian, a flexible and cost-effective method that combines mixed-precision quantization with attribute-discriminative pruning for training-free 3D Gaussian compression. FlexGaussian eliminates the need for retraining and adapts easily to diverse compression targets. Evaluation results show that FlexGaussian achieves up to 96.4% compression while maintaining high rendering quality (<1 dB drop in PSNR), and is deployable on mobile devices. FlexGaussian delivers high compression ratios within seconds, being 1.7-2.1x faster than state-of-the-art training-free methods and 10-100x faster than training-involved approaches. The code is being prepared and will be released soon at: https://github.com/Supercomputing-System-AI-Lab/FlexGaussian

cs.CV↗

arXiv2025

Universal Checkpointing: A Flexible and Efficient Distributed Checkpointing System for Large-Scale DNN Training with Reconfigurable Parallelis

Deep neural network (DNN) training continues to scale rapidly in terms of model size, data volume, and sequence length, to the point where multiple machines are required to fit large models for training. Different distributed and parallel training strategies have been developed to support large-scale DNN training by partitioning the training state across GPUs. However, existing DNN training systems provide very limited support for reconfiguring parallelism strategies in the middle of the training via checkpointing. This limitation arises because distributed checkpoints are tightly coupled to specific model parallelism and hardware configurations, preventing large-scale training jobs from efficiently adapting to hardware failures or resource elasticity. This paper presents Universal Checkpointing (UCP), a novel checkpointing system that enables flexible and efficient DNN training with reconfigurable parallelism. UCP overcomes challenges in existing systems by decoupling checkpoint structure from parallel training strategies and hardware configurations. In addition, we present a pattern-based reconfiguration pipeline that enables automatic, flexible, and efficient mapping of checkpoint state to various parallelism strategies. Evaluation on a range of DNN models, including state-of-the-art dense and sparse LLMs, shows that UCP enables reconfiguration for a broader set of widely used parallelism strategies than existing solutions while adding negligible reconfiguration cost. UCP has been successfully employed in real LLM training workloads, greatly enhancing their flexibility and resilience to dynamic hardware environments.

cs.DC↗

arXiv2025

VQ-LLM: High-performance Code Generation for Vector Quantization Augmented LLM Inference

In this work, we design and implement VQ-LLM, an efficient fused Vector Quantization (VQ) kernel generation framework. We first introduce a software abstraction called codebook cache to optimize codebook access efficiency and support the integration of VQ with various computations. The codebook cache adaptively stores different entries across the GPU's memory hierarchy, including off-chip global memory, on-chip shared memory, and registers. Centered around the codebook cache, we design an efficient computation engine that optimizes memory traffic during computations involving codebooks. This compute engine adopts the codebook-centric dataflow and fusion optimizations. Additionally, we provide adaptive heuristics to tailor parameter selection in our optimizations to diverse VQ configurations. Our optimizations achieve an average latency reduction of 46.13% compared to unoptimized versions. Compared to existing open-source implementations, our methods decrease latency by 64.36% to 99.1%. A final comparison with state-of-the-art element-wise quantization methods like AWQ and KVQuant shows that our VQ-LLM is practically viable, achieving latencies close or even better latencies to those at equivalent bit-widths, potentially offering greater accuracy.

cs.DC↗

arXiv2025

MiniKV: Pushing the Limits of LLM Inference via 2-Bit Layer-Discriminative KV Cache

How to efficiently serve LLMs in practice has become exceptionally challenging due to their prohibitive memory and computation requirements. In this study, we investigate optimizing the KV cache, whose memory footprint poses a critical bottleneck in LLM inference, especially when dealing with long context tasks. To tackle the challenge, we introduce MiniKV, a KV cache optimization method that simultaneously preserves long context task accuracy while significantly reducing KV cache size via a novel 2-bit layer-discriminative KV cache. More importantly, we develop specialized CUDA kernels to make MiniKV compatible with FlashAttention. Experiments on a wide range of long context tasks show that MiniKV effectively achieves 86% KV cache compression ratio while recovering over 98.5% of accuracy, outperforming state-of-the-art methods while achieving excellent measured system performance improvements.

cs.CL↗

arXiv2025

MedCite: Can Language Models Generate Verifiable Text for Medicine?

Existing LLM-based medical question-answering systems lack citation generation and evaluation capabilities, raising concerns about their adoption in practice. In this work, we introduce \name, the first end-to-end framework that facilitates the design and evaluation of citation generation with LLMs for medical tasks. Meanwhile, we introduce a novel multi-pass retrieval-citation method that generates high-quality citations. Our evaluation highlights the challenges and opportunities of citation generation for medical tasks, while identifying important design choices that have a significant impact on the final citation quality. Our proposed method achieves superior citation precision and recall improvements compared to strong baseline methods, and we show that evaluation results correlate well with annotation results from professional experts.

cs.CL↗

arXiv2025

VecFlow: A High-Performance Vector Data Management System for Filtered-Search on GPUs

Vector search and database systems have become a keystone component in many AI applications. While many prior research has investigated how to accelerate the performance of generic vector search, emerging AI applications require running more sophisticated vector queries efficiently, such as vector search with attribute filters. Unfortunately, recent filtered-ANNS solutions are primarily designed for CPUs, with few exploration and limited performance of filtered-ANNS that take advantage of the massive parallelism offered by GPUs. In this paper, we present VecFlow, a novel high-performance vector filtered search system that achieves unprecedented high throughput and recall while obtaining low latency for filtered-ANNS on GPUs. We propose a novel label-centric indexing and search algorithm that significantly improves the selectivity of ANNS with filters. In addition to algorithmic level optimization, we provide architectural-aware optimization for VecFlow's functional modules, effectively supporting both small batch and large batch queries, and single-label and multi-label query processing. Experimental results on NVIDIA A100 GPU over several public available datasets validate that VecFlow achieves 5 million QPS for recall 90%, outperforming state-of-the-art CPU-based solutions such as Filtered-DiskANN by up to 135 times. Alternatively, VecFlow can easily extend its support to high recall 99% regime, whereas strong GPU-based baselines plateau at around 80% recall. The source code is available at https://github.com/Supercomputing-System-AI-Lab/VecFlow.

cs.DB↗