arXiv subjects

Deyu Cao

Publications and source records attributed to Deyu Cao.

5 recordsLinked to original sources

arXiv2026

KAT-Coder-V2.5 Technical Report

We present KAT-Coder-V2.5, a coding-focused agentic model trained to act autonomously inside real, executable repositories rather than as a single-turn code generator. Its capability is bottlenecked less by model scale than by the scarcity of reproducible environments, verifiable rewards, and high-value trajectories, which we address with an end-to-end agentic post-training framework. AutoBuilder reconstructs multilingual repositories into sandboxed environments with fail-to-pass and pass-to-pass verification at scale, from which we regenerate self-contained task specifications, recover near-miss trajectories, and distill supervision through process-aware filtering, while KwaiClawEnv synthesizes large-scale tool-use trajectories from executable services and real task seeds. We further scale reinforcement learning with harness randomization, a reliability-hardened sandbox, an asymmetric actor--critic PPO with hindsight-augmented value estimation, and a harness-oriented reward framework, and unify SWE, Agent-Claw, and WebCoding experts via Multi-Teacher On-Policy Distillation. Across six software-engineering and agentic benchmarks, KAT-Coder-V2.5 delivers the best agentic tool-use result on PinchBench and ranks second only to the frontier Opus 4.8 on repository-level software engineering. Our service is available at https://streamlake.com/product/kat-coder.

cs.SE↗

arXiv2026

Sliced-Wasserstein Distribution Alignment Loss Improves the Ultra-Low-Bit Quantization of Large Language Models

The benefits of most large language models come with steep and often hidden economic and environmental costs due to their resource usage inefficiency during deployment. Model quantization improves energy and memory efficiency through representing model parameters by lower-precision values. However, compression below 4-bits often distorts activation distributions and degrades performance. We address this challenge by introducing a sliced Wasserstein loss function for distribution-aware calibration in ultra-low-bit post-training quantization. The proposed loss aligns the output distributions of full-precision and quantized models under random linear projections, complementing standard mean-squared error loss without adding any computational overhead during inference. Our proposed loss function can be incorporated with any post-training quantization framework that has a retraining component. We demonstrate the performance gains of our proposed model by incorporating it with two frontier methods known as OmniQuant and TesseraQ. Compared to these two baselines, the proposed loss consistently improves both perplexity and downstream task accuracy across multiple ultra-low-bit settings. Our proposed loss function recovers 4.12-20.37% of the OmniQuant's lost accuracy on the language model LLaMA-2-7B, 0.93-7.65% on OPT-6.7B, and 2.26-6.20% on LLaMA-2-13B. TesseraQ's accuracy degradation is recovered by 3.63-7.63% in relative terms when augmented by our proposed loss function. Taken together, these results demonstrate that distributional alignment provides a simple yet effective performance boost that can push the limits of frontier quantization methods. Our method is available on GitHub to facilitate future progress in ultra-low-bit quantization.

cs.LG↗

arXiv2025

Flows and Diffusions on the Neural Manifold

Diffusion and flow-based generative models have achieved remarkable success in domains such as image synthesis, video generation, and natural language modeling. In this work, we extend these advances to weight space learning by leveraging recent techniques to incorporate structural priors derived from optimization dynamics. Central to our approach is modeling the trajectory induced by gradient descent as a trajectory inference problem. We unify several trajectory inference techniques towards matching a gradient flow, providing a theoretical framework for treating optimization paths as inductive bias. We further explore architectural and algorithmic choices, including reward fine-tuning by adjoint matching, the use of autoencoders for latent weight representation, conditioning on task-specific context data, and adopting informative source distributions such as Kaiming uniform. Experiments demonstrate that our method matches or surpasses baselines in generating in-distribution weights, improves initialization for downstream training, and supports fine-tuning to enhance performance. Finally, we illustrate a practical application in safety-critical systems: detecting harmful covariate shifts, where our method outperforms the closest comparable baseline.

cs.LG↗

arXiv2025

Enhancing Ultra-Low-Bit Quantization of Large Language Models Through Saliency-Aware Partial Retraining

The growing use of large language models has raised environmental and economic concerns about their intensity of resource usage during inference. Serving these models to each user requires substantial energy and water for cooling. Model compression techniques like quantization can shrink large language models and make them more resource efficient at the cost of potential performance degradation. Quantization methods compress model size through replacing their high-precision parameters by quantized values of lower precision. Among existing methods, the ApiQ method achieves superior accuracy preservation at minimal memory and time overhead. We investigate two ideas to extend performance in ultra-low-bit quantization beyond ApiQ's level. First, we look into combining existing quantization-aware training techniques with ApiQ's partial training. We show that this does not outperform the baseline ApiQ method with limited training data and frozen weights. This leads to two key insights: (1) The substantial representational capacity that is gained through full retraining is unlikely to be feasible through partial training. (2) This gain may depend on using a large and diverse dataset in quantization-aware training. Second, through a novel approach informed by the two insights, we propose an ultra-low-bit quantization method that builds upon ApiQ and extends its performance without the need for full retraining. This publicly available method relies on a saliency-aware regularization term that prioritizes preserving the most impactful parameters during quantization. Our experiments on LLaMA 7B and 13B benchmarks demonstrate that our method reduces the ApiQ's accuracy degradation by 10.85% and 7.54% respectively. A Python implementation of the proposed quantization method is publicly available on GitHub https://github.com/TokuyuSou/ULB-SAPR.

cs.LG↗

arXiv2025

Flow to Learn: Flow Matching on Neural Network Parameters

Foundational language models show a remarkable ability to learn new concepts during inference via context data. However, similar work for images lag behind. To address this challenge, we introduce FLoWN, a flow matching model that learns to generate neural network parameters for different tasks. Our approach models the flow on latent space, while conditioning the process on context data. Experiments verify that FLoWN attains various desiderata for a meta-learning model. In addition, it matches or exceeds baselines on in-distribution tasks, provides better initializations for classifier training, and is performant on out-of-distribution few-shot tasks while having a fine-tuning mechanism to improve performance.

cs.LG↗