Skip to content
View selimfedakar's full-sized avatar
5 followers · 4 following

Achievements

Achievement: Pull Sharkx3Achievement: YOLO

Achievements

Achievement: Pull Sharkx3Achievement: YOLO

Block or report selimfedakar

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
selimfedakar/README.md

Python PyTorch NumPy CUDA Metal MLX TypeScript

LLMs and ML systems from scratch, measured rather than assumed.

  • co-founder of EXCAR, an in-car voice AI running on real devices
  • scratchbench: from-scratch neural nets, benchmarked on real hardware
  • handmade-llm: a language model built and trained end to end in plain Python, no frameworks
  • upstream fixes across the ML open-source ecosystem, every bug proven with a reproducible test

I care about high-performance computing: arithmetic intensity, memory bandwidth, and why the profiler disagrees with the paper.

📍 Los Angeles · a.selimfedakar@gmail.com · selimfedakar.com

Pinned Loading

  1. My-Neural-Networks-Projects My-Neural-Networks-Projects Public

    A collection of my neural network implementations, starting from scratch with an autograd engine.

    Python

  2. My_Stanford_CS336_Language_Modeling My_Stanford_CS336_Language_Modeling Public

    A collection of my implementation of Stanford CS336: Language Modeling from-scratch. Deep-dives into MFU, tensor memory optimization (strides/contiguous), and SOTA architectural consensus: RoPE, Sw…

    Python

  3. scratchbench scratchbench Public

    Can the model actually implement it? ML machinery - attention masks, KV-caches, quantization, a Triton kernel and a Metal one - written from a specification and graded by hidden pytest suites. On a…

    Python

  4. handmade-llm handmade-llm Public

    Train a real LLM from scratch on the Apple Silicon Mac you already own, byte-level BPE, a transformer in MLX, the training loop, then ship it to your iPhone. No CUDA, no cloud bill, no discrete GP…

    Python