Meta/Facebook seeking full-time research scientists in Architecture, HPC and ML Systems

If interested, please message me on LinkedIn with your resume.

Meta, formerly known as Facebook, is seeking full-time research scientists and software engineers in Computer Science, specifically with experience in architecture, HPC, and ML systems:?

• GPU/ASIC architecture
• GPU/ASIC-based ML kernel development and optimization (e.g. CUDA, ROCm)?
• CPU-based threading models for x86 and ARM (e.g., OpenMP, Pthreads)
• HPC communication libraries (e.g., NCCL, RCCL, UCC, MPI)
• Numeric libraries (e.g., low precision arithmetic, quantizations, mixed precision ops)
• High performance computing (HPC)
• Distributed systems for hyperscale large language model (LLM) training
• ML compiler and ML distributed technologies

These fields largely correspond to the research communities of ISCA, MLSys, Supercomputing (ICS/SC), and AI-related parts of SOSP/OSDI/ASPLOS/NSDI as well as NeurIPS, ICML, KDD, etc.?

The team work spans across both software and hardware. On the software side, among other things, we directly contribute to LLama-3, DLRM, and open-source projects like PyTorch FSDP and PyTorch2 compiler. On the hardware side, we directly contribute to the recently announced MTIA accelerator and the GrandTeton platform. In addition to numerous top-tier publications in related fields, our distinctive strength lies in the fact that our work is directly incorporated into production systems that impact billions of users across our products, including Instagram, Facebook, and meta.ai.

Ideal candidates should hold a Ph.D. and boast a strong publication record in top-tier conferences, including those mentioned earlier. While we excel in research and publication, our core emphasis lies in developing cutting-edge hardware and software innovations that run efficiently at scale and are integrated into production systems that connect billions of people world-wide on Meta platforms. It is important to note that we do not consider research papers as our end products, and our roles require extensive hands-on engineering work.