LLM training in simple, raw C/CUDA
Role in this project:
ML Engineer Contributions:6 reviews, 41 PRs, 33 comments in 1 month
Contributions summary:Chris primarily contributes to CUDA kernels for training a language model. Their work focuses on optimizing and clarifying the code, including removing unused functions, removing templates to clarify data types, and adding packing for gelu forwards kernel. They also refactor existing kernels and add support for calculating layernorm activations during the backward pass.
Crowdsourced List of Upcoming Hackathons in Toronto Organized by Date and Application Deadline
Contributions:102 commits, 18 PRs, 98 pushes in 6 months
reactcrowdsourceddeadlineagile-developmentorganized