AMDAI Framework Eng.

任职要求

Benefits offered are described:  AMD benefits at a glance.
 
AMD does not accept unsolicited resumes from headhunters, recruitment agencies, or fee-based recruitment services. AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition, mental or physical disability, national origin, race, religion, political and/or third-party affiliation, sex, pregnancy, sexual orientation, …

工作职责

THE ROLE: 
We are looking for a dynamic, upbeat software engineer to join our growing team. Your work will focus on building robust, efficient software components that enable high-performance execution of large language models and multimodal models across multi-GPU systems. You’ll collaborate with internal GPU library teams and open-source maintainers to implement features that improve throughput, latency, and scalability. This role emphasizes full-stack development within AI inference systems, with a strong focus on model behavior and framework integration.
 
THE PERSON: 
A motivated early-career software engineer with solid foundational skills in Python and/or C++ in Linux environments. The ideal candidate has hands-on experience or strong academic exposure to deep learning systems, understands LLM and multimodal model architectures, and is eager to write production-quality code that balances functionality, correctness, and performance. 
 
KEY RESPONSIBILITIES:  
Deep Learning & LLM Framework Optimization: Experience with optimizing major DL/LLM frameworks (PyTorch, vLLM, SGLang) for AMD GPUs and contribute improvements upstream. Model-Aware Implementation: Build features that interact closely with LLMs and multimodal architectures (e.g., Llama, Qwen-VL, Wan), requiring understanding of attention mechanisms, cross-modal fusion, KV caching, and quantization. Performance-Conscious Coding: Write efficient, scalable code while considering memory usage, concurrency, and bottlenecks in multi-GPU environments. Profiling: Use profiling tools to evaluate the impact of your changes, identify regressions, and validate performance improvements as part of the development cycle. End-to-End Performance Engineering: Perform comprehensive profiling to identify bottlenecks and implement system, memory, and communication optimizations across multi-GPU and multi-node setups. Compiler & Pipeline Acceleration: Leverage compiler technologies and graph compilers to enhance the full deep learning and inference pipeline. Research & Advanced Techniques: Prototype and integrate emerging optimization methods such as speculative decoding and weight-only quantization into production systems. Cross-Team & Open-Source Collaboration: Collaborate with internal GPU library teams and open-source maintainers to align improvements and ensure seamless upstream integration. Software Engineering Excellence: Apply robust engineering practices to deliver maintainable, reliable, and production-quality performance optimizations. MANDATORY EXPERIENCE:  
Software Engineering Skills: Familiarity in Python. Familiarity with C++ or async programming is a plus. Understanding of LLM or multimodal model concepts: Knowledge of transformer architectures, attention mechanisms, vision-language alignment, and inference pipelines (e.g., image + text input handling). Have theoretical grounding in Transformer/Attention/MoE/KV Cache, and quantization (FP8/FP4). Linux development environment: Comfortable using command-line tools, Git, and standard debugging/profiling utilities. End-to-End LLM Performance Engineering: Experience with profiling and diagnosing compute, memory, and communication bottlenecks across multi-GPU and multi-node environments. Software Engineering Excellence & Community Contribution is a plus: Solid Python/C++ coding skills and experience debugging and testing practices, proven ability to deliver maintainable performance-critical software, and a track record of open-source contributions with strong self-motivation. GPU Kernel Development & Optimization is a plus: Knowlege of high-performance GPU kernels tuning for AMD GPUs using HIP, CUDA, ASM, and tools like CK, CUTLASS, and Triton. Compiler & System-Level Optimization is a plus: Foundational knowledge of LLVM, ROCm, and compiler-driven techniques for improving kernel and system performance. Model Architectures & Optimization Expertise: Knowledge with multimodal models (e.g., Qwen-VL, Qwen-Image-Edit, Wan) or diffusion-based generative models. Development Skills: Exposure to GPU computing (ROCm, CUDA) or performance profiling tools (e.g., PyTorch Profiler). Distributed Systems Experience: Experience with distributed inference for large-scale models (e.g., Tensor Parallel, Pipeline Parallel). ACADEMIC & PREFERRED QUALIFICATIONS:  
Bachelor’s in Computer Science, Computer Engineering, Electrical Engineering, or a related field.  
#LI-EH1