高级软件工程师
Sr Software Engineer
关于 Positron AI Positron AI 专注于开发自定义硬件系统以加速人工智能推理。这些推理系统比传统的 GPU 基于系统提供了显著的性能和效率增长，在美元和瓦特均性能方面都带来优势.Positron 存在于创造世界上最好的人工智能论断系统。角色概述高级软件工程师 机器学习系统 & 高性能 LLM 推理 我们正在寻找一名高级软件技术工程师，为开发高性能定制软件做出贡献，该软件能够在其设备上执行开源大型语言模型 (LLMs). 该设备利用 FPGAs 和 x86CPU 的组合来加速基于变压器的模型。该软件堆主要以现代 C++ (C++17/20) 编写，并严重依赖于模板，SIMD 优化和高效的平行计算技术。关键责任 ·设计和实现针对自定义硬件 LLM 的高性能推断软件。·开发和优化使用 SIMD 指令，线程和内存等级有效的 C++基础库。·与 FPGA 和系统工程师密切合作，确保 x86CPUs 和 FPGAs 之间的数据流动和计算卸载。·通过低级优化，包括向量化，缓存效率和硬件知情的规划。· 贡献性能配置工具和方法来分析指令和数据流水平上的执行瓶。·应用 NUMA 知情的内存管理技术来优化大规模推理工作负载的内存访问模式。·实施 ML 系统层面优化，如代币流媒体，KV 缓存优化以及变压器执行的高效批量。·与 ML 研究人员和软件工程师合作整合模型定量化技术，稀疏优化和混合精度执行。·确保所有代码贡献包括单元，性能，接受和回归测试作为持续集成基础开发过程的一部分。要求资格 · 7年以上在 C++软件开发方面的专业经验，重点放在性能关键应用上。·对 C++模板和现代内存管理的强有力的理解。·与 SIMD 编程 （AVX-512,SSE 或相等）和基于内部矢量化的实践经验。· 高性能计算机的经验。
查看英文原文
About Positron AI
Positron AI specializes in developing custom hardware systems to accelerate AI inference. These inference systems offer significant performance and efficiency gains over traditional GPU-based systems, delivering advantages in both performance per dollar and performance per watt. Positron exists to create the world's best AI inference systems.
Role Overview
Senior Software Engineer – Machine Learning Systems & High-Performance LLM Inference
We are seeking a Senior Software Engineer to contribute to the development of high-performance software that powers execution of open-source large language models (LLMs) on our custom appliance. This appliance leverages a combination of FPGAs and x86 CPUs to accelerate transformer-based models. The software stack is written primarily in modern C++ (C++17/20) and heavily relies on templates, SIMD optimizations, and efficient parallel computing techniques.
Key Responsibilities
- Design and implement high-performance inference software for LLMs on custom hardware.
- Develop and optimize C++-based libraries that efficiently utilize SIMD instructions, threading, and memory hierarchy.
- Work closely with FPGA and systems engineers to ensure efficient data movement and computational offloading between x86 CPUs and FPGAs.
- Optimize model execution via low-level optimizations, including vectorization, cache efficiency, and hardware-aware scheduling.
- Contribute to performance profiling tools and methodologies to analyze execution bottlenecks at the instruction and data flow levels.
- Apply NUMA-aware memory management techniques to optimize memory access patterns for large-scale inference workloads.
- Implement ML system-level optimizations such as token streaming, KV cache optimizations, and efficient batching for transformer execution.
- Collaborate with ML researchers and software engineers to integrate model quantization techniques, sparsity optimizations, and mixed-precision execution.
- Ensure all code contributions include unit, performance, acceptance, and regression tests as part of a continuous integration-based development process.
Required Qualifications
- 7+ years of professional experience in C++ software development, with a focus on performance-critical applications.
- Strong understanding of C++ templates and modern memory management.
- Hands-on experience with SIMD programming (AVX-512, SSE, or equivalent) and intrinsics-based vectorization.
- Experience in high-performance computing (HPC), numerical computing, or ML inference optimization.
- Experience with ML model execution optimizations, including efficient tensor computations and memory access patterns.
- Knowledge of multi-threading, NUMA architectures, and low-level CPU optimization.
- Proficiency with systems-level software development, profiling tools (perfetto, VTune, Valgrind), and benchmarking.
- Experience working with hardware accelerators (FPGAs, GPUs, or custom ASICs) and designing efficient software-hardware interfaces.
Preferred Qualifications
- Familiarity with LLVM/Clang or GCC compiler optimizations.
- Experience in LLM quantization, sparsity optimizations, and mixed-precision computation.
- Knowledge of distributed inference techniques and networking optimizations.
- Understanding of graph partitioning and execution scheduling for large-scale ML models.
Leveling & Scope
While this role is currently posted at a specific level, we are a growth-oriented organization and are open to hiring at a more senior level for the right candidate. Please note that this job description serves as a focused but generalized overview of the role; specific responsibilities and impact expectations will be tailored to the experience and seniority of the final hire.
Why Join Us?
- Work on a cutting-edge ML inference platform that redefines performance and efficiency for LLMs.
- Tackle challenging low-level performance engineering problems in AI and HPC.
- Collaborate with a team of hardware, software, and ML experts building an industry-first product.
- Opportunity to contribute to and shape the future of open-source AI inference software.
Compensation and Benefits
The base salary range for this role is $150,000 – $250,000.
Please note that the figures provided represent the base salary range only and do not include other elements of our total compensation package, equity, or comprehensive benefits.
At Positron AI, we value the unique expertise each candidate brings. While the range above reflects our typical expectation for the position, we reserve the flexibility to exceed this range for candidates whose specialized skills, significant experience, or unique qualifications fall outside the standard scope of the role. Final offers are determined based on a variety of factors, including internal equity, and individual impact.
Equal Opportunity Employer. If you’re excited about the role but don’t meet every bullet, we’d still love to hear from you.
Originally posted on Himalayas