arXiv subjects

Yibo He

Publications and source records attributed to Yibo He.

7 recordsLinked to original sources

arXiv2026

Kimi K3: Open Frontier Intelligence

We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token context window. Kimi K3 is built on Kimi Delta Attention and Attention Residuals, which improve information flow across sequence length and model depth. Together with Stable LatentMoE, which effectively activates 16 of 896 routed experts per token, and refined training and data recipes, these advances yield an approximately 2.5x improvement in overall scaling efficiency over Kimi K2. Post-training highlights reinforcement learning across general, agentic, and coding domains and multiple reasoning-effort levels, enabling compositional generalization and robust long-horizon execution. At 2.8T scale, Kimi K3 is supported by infrastructure advances in multiple areas: algorithm-system co-design for KDA, perfectly balanced expert-parallel training with efficient memory management, million-token agentic RL with persistent rollout and sandbox states, and deployment innovations. Extensive evaluations show that Kimi K3 achieves frontier-level performance across long-horizon coding, agentic, knowledge, reasoning, and vision tasks. While its overall performance still trails the most powerful proprietary models, namely Claude Fable 5 and GPT-5.6 Sol, Kimi K3 consistently outperforms other open and proprietary models evaluated in our suite. We release the full Kimi K3 model weights to facilitate future research and accelerate the broader deployment and adoption of frontier intelligence.

cs.CL↗

arXiv2026

YouZhi: Towards High-Concurrency Financial LLMs via Adaptive GQA-to-MLA Transition

Large language models (LLMs) drive significant financial innovations, yet their high-concurrency deployment is severely bottlenecked by KV cache memory overhead, which inflates infrastructure costs and throttles scalability. To address this, we propose YouZhi-LLM, a highly efficient financial LLM empowered by a comprehensive structural transition and training pipeline natively built on the Huawei Ascend ecosystem. At its algorithmic core, YouZhi-LLM features a layer-adaptive GQA-to-MLA transition framework that dynamically assigns per-layer FreqFold sizes, maximizing KV-cache compression while minimizing perplexity degradation. To recover representation capacity and inject domain expertise, the Ascend-based training pipeline seamlessly integrates generalized knowledge distillation with financial-specific supervised fine-tuning. Evaluations demonstrate the superiority of this systematic approach, with the adaptive transition reducing perplexity degradation by up to 35% over uniform baselines. Crucially, when evaluated on Ascend NPUs via vLLM-Ascend, the massive KV-cache reduction translates directly into deployment efficiency. Compared to their respective base models, YouZhi-7B yields a 12.3% improvement in average financial benchmark score alongside a 2.69$\times$ increase in maximum concurrency; similarly, YouZhi-14B achieves a 7.0% accuracy gain and a 2.43$\times$ concurrency boost, establishing a new paradigm for cost-effective, high-throughput financial inference.

cs.CL↗

arXiv2026

Synthesizing File-Level Data for Unit Test Generation with Chain-of-Thoughts via Self-Debugging

Automatic unit test (UT) generation is essential for software quality assurance, but existing approaches--including symbolic execution, search-based approaches, and recent LLM-based generators--struggle to produce human-quality tests with correct, meaningful assertions and reliable chain-of-thought (CoT) explanations. We identify a gap in UT training data: repository-mined tests lack developer CoTs, while LLM-distilled CoTs are often incorrect or incomplete. To address this issue, we propose a novel data-distillation approach that uses self-debugging to produce high-quality UT training examples paired with faithful CoTs. Our approach combines (1) guided test repair, a heuristic loop (error-, failure-, and coverage-focused steps) that asks the used model to diagnose and iteratively fix generated tests, and (2) CoT compression, which compacts original and debugging CoTs into concise explanations that directly justify correct tests. We apply this pipeline to a large corpus of open-source projects to construct a dataset of 74,518 high-quality examples, and then use it for supervised fine-tuning of a base model. An empirical evaluation shows that the fine-tuned model achieves high UT generation effectiveness: it attains a pass rate of 36.17% on test assertions, a branch coverage of 43.90%, and a mutation score of 88.66%, substantially higher than state-of-the-art commercial models like o4-mini.

cs.SE↗

arXiv2026

Kimi K2.5: Visual Agentic Intelligence

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

cs.CL↗

arXiv2025

SimdBench: Benchmarking Large Language Models for SIMD-Intrinsic Code Generation

SIMD (Single Instruction Multiple Data) instructions and their compiler intrinsics are widely supported by modern processors to accelerate performance-critical tasks. SIMD intrinsic programming, a trade-off between coding productivity and high performance, is widely used in the development of mainstream performance-critical libraries and daily computing tasks. Large Language Models (LLMs), which have demonstrated strong and comprehensive capabilities in code generation, show promise in assisting programmers with the challenges of SIMD intrinsic programming. However, existing code-generation benchmarks focus on only scalar code, and it is unclear how LLMs perform in generating vectorized code using SIMD intrinsics. To fill this gap, we propose SimdBench, the first code benchmark specifically designed for SIMD-intrinsic code generation, comprising 136 carefully crafted tasks and targeting five representative SIMD intrinsics: SSE (x86 Streaming SIMD Extension), AVX (x86 Advanced Vector Extension), Neon (ARM Advanced SIMD Extension), SVE (ARM Scalable Vector Extension), and RVV (RISC-V Vector Extension). We conduct a systematic evaluation (measuring both correctness and performance) of 18 representative LLMs on SimdBench, resulting in a series of novel and insightful findings. Our evaluation results demonstrate that LLMs exhibit a universal decrease in pass@k during SIMD-intrinsic code generation compared to scalar-code generation. Our in-depth analysis highlights promising directions for the further advancement of LLMs in the challenging domain of SIMD-intrinsic code generation. SimdBench is fully open source at https://anonymous.4open.science/r/SimdBench-1B3F/ to benefit the broader research community.

cs.SE↗

arXiv2025

RVISmith: Fuzzing Compilers for RVV Intrinsics

Modern processors are equipped with single instruction multiple data (SIMD) instructions for fine-grained data parallelism. Compiler auto-vectorization techniques that target SIMD instructions face performance limitations due to insufficient information available at compile time, requiring programmers to manually manipulate SIMD instructions. SIMD intrinsics, a type of built-in function provided by modern compilers, enable programmers to manipulate SIMD instructions within high-level programming languages. Bugs in compilers for SIMD intrinsics can introduce potential threats to software security, producing unintended calculation results, data loss, program crashes, etc. To detect bugs in compilers for SIMD intrinsics, we propose RVISmith, a randomized fuzzer that generates well-defined C programs that include various invocation sequences of RVV (RISC-V Vector Extension) intrinsics. We design RVISmith to achieve the following objectives: (i) achieving high intrinsic coverage, (ii) improving sequence variety, and (iii) without known undefined behaviors. We implement RVISmith based on the ratified RVV intrinsic specification and evaluate our approach with three modern compilers: GCC, LLVM, and XuanTie. Experimental results show that RVISmith achieves 11.5 times higher intrinsic coverage than the state-of-the-art fuzzer for RVV intrinsics. By differential testing that compares results across different compilers, optimizations, and equivalent programs, we detect and report 13 previously unknown bugs of the three compilers under test to date. Of these bugs, 10 are confirmed and another 3 are fixed by the compiler developers.

cs.CR↗

arXiv2024

InfiBench: Evaluating the Question-Answering Capabilities of Code Large Language Models

Large Language Models for code (code LLMs) have witnessed tremendous progress in recent years. With the rapid development of code LLMs, many popular evaluation benchmarks, such as HumanEval, DS-1000, and MBPP, have emerged to measure the performance of code LLMs with a particular focus on code generation tasks. However, they are insufficient to cover the full range of expected capabilities of code LLMs, which span beyond code generation to answering diverse coding-related questions. To fill this gap, we propose InfiBench, the first large-scale freeform question-answering (QA) benchmark for code to our knowledge, comprising 234 carefully selected high-quality Stack Overflow questions that span across 15 programming languages. InfiBench uses four types of model-free automatic metrics to evaluate response correctness where domain experts carefully concretize the criterion for each question. We conduct a systematic evaluation for over 100 latest code LLMs on InfiBench, leading to a series of novel and insightful findings. Our detailed analyses showcase potential directions for further advancement of code LLMs. InfiBench is fully open source at https://infi-coder.github.io/infibench and continuously expanding to foster more scientific and systematic practices for code LLM evaluation.

cs.SE↗