NVIDIA is seeking to discover and innovate new low-precision and sparsity recipes in the pretraining setting to develop next-generation software for novel hardware features on current GPUs
Requirements
- Proficient in Python
- Experience with PyTorch or similar framework
- Solid foundation in LLM pre training, post training, or generation
- Proficient in the math of machine learning
- Proficient in precision and numerics for ML
- Familiarity with FP8 and MX formats for training
Responsibilities
- Keep abreast on quantized LLM training research
- Build robust and reproducible training recipes
- Collaborate closely with hardware, software, and research teams to assess and adopt deep learning algorithmic advancements in quantization
- Work with production SW teams to realize recipes in production workflows
Other
- PhD or M.S. degree (or equivalent experience) in Computer Science or a related field, and 5+ years of relevant software engineering experience
- Strong written and oral communication skills
- Ability to work in a diverse environment
- Must be eligible to work in the country without sponsorship
- Travel may be required
