Testing out the huggingface version for the gpt-oss-20b locally on consumer hardware

Regularization to improve duplication penalty loss

How Triton Compiler Works Under the Hood!

Enough MLIR to be dangerous - how Triton uses MLIR passes to progressively lower IR

Improve model to reduce duplicates and improve training performance

Exploring a simple transformer model for sequence modelling in recommender systems

What happens when triton.compile is called in the frontend?

Missing tutorial on how triton program gets converted to cuda kernels under the hood

Benchmarking our own GPT2 model against Huggingface GPT2 model

Writing GPT2 from scratch and assigning weights from pre-trained Huggingface model