Research Engineer, Algorithms for LLM Inference Hardware
Develop LLM inference algorithms for Normal Computing’s in-memory AI hardware. Collaborate with hardware and architecture teams to co-design numerical methods, benchmarks, and prototypes.
Apraksts
- Develop algorithms for LLM inference workloads on in-memory hardware.
- Partner with hardware and architecture teams on software and hardware co-design.
- Create numerical methods that use thermal noise and analog dynamics.
- Build evaluation frameworks and benchmarks for simulation and physical hardware.
- Turn insights about model workloads into hardware design constraints and opportunities.
- Prototype and refine algorithms as hardware progresses from simulation to silicon.
- Improve performance at both the gate and algorithm levels.
- Test algorithms with simulators, FPGA prototypes, and silicon as it becomes available.
Prasības
- Strong knowledge of large-model inference, including attention mechanisms, KV cache, long-context decoding, and memory-bandwidth limits.
- Experience improving inference through techniques such as quantization, sparsity, kernel fusion, or memory-efficient attention.
- Familiarity with stochastic systems, probabilistic methods, numerical analysis, or analog computation.
- Experience implementing algorithms close to hardware rather than relying only on high-level frameworks.
- Ability to reason from first principles about new computing substrates.
- Demonstrated ability to take ideas from theory through to implementation on real hardware.
- Strong Python skills and proficiency in at least one systems programming language.
- Ability to collaborate across hardware, architecture, and software teams.
- A PhD in machine learning, applied mathematics, physics, electrical engineering, or a related field is a plus.
- Exposure to analog or mixed-signal systems, in-memory computing, or non-von-Neumann architectures is a plus.
- Experience working on hardware that was still being developed when you joined is a plus.
- Publications or open-source contributions in efficient inference, stochastic algorithms, or novel computing are a plus.
Priekšrocības
- An inclusive workplace with a commitment to diversity.
- Reasonable accommodations for people with disabilities.
- Equal employment opportunity protections.
Saistītās vakances
Nav atrasts neviens darbs
Mēģiniet pielāgot vai notīrīt filtrus, lai atrastu vairāk iespēju.