I also like to document my open-sourced work, many of which are rather random. It reminds me of the good work I have done, no matter how trivial

Basic slides detailing some work

vllm-project/tpu-inference#1644 [AWQ][MoE] Support AWQ MoE

vllm-project/tpu-inference#1729 Support DeepSeek-V2-Lite-Chat-FP8

vllm-project/tpu-inference#1388 Support FP8 quantization

vllm-project/tpu-inference#1911 [Kernel][GMM] Support zero point asymmetric quantization [unmerged. reference only]

vllm-project/tpu-inference#1440 Handle unaligned FP8 block quantization edge case [DeepSeek]

vllm-project/tpu-inference#1659 Properly register quantization configs

vllm-project/tpu-inference#1472 Support MoE sigmoid scoring

vllm-project/tpu-inference#1656 Update dequantization dtype to match comments

vllm-project/tpu-inference#1454 Remove dead code after upstream removal

vllm-project/vllm#33501 Fix DeepSeek RoPE initialization error

llvm/torch-mlir#4228 [OnnxToTorch] Casting float to integer should round to nearest for pow with int result type

marin-community/levanter#974 Initial implementation of power scheduler

pytorch/pytorch#163616 Update doc for torch.index_select, which can accept IntTensor as a parameter

pytorch/tutorials#3844 Update torch.compile triton.autotune note

jax-ml/jax#35492 [Pallas TPU] Incorrect result for jnp.dot with uint4 rhs and bfloat16 lhs

jax-ml/jax#34620 [Pallas TPU] jnp.argmax returns last index instead of first index on ties

jax-ml/jax#34674 [Mosaic] Update logistic lowering to correctly handle BF16 inputs

pytorch/pytorch#164666 sparse.mm(CSR x dense) backward fails on CUDA for bfloat16 but CSR @ dense backward works

mintisan/awesome-kan#139 Add LinearKAN

AlphaGPU/leetgpu-challenges#90 Script to generate starter code for GPU challenges

TheOdinProject/javascript-exercises#370 Update fibonacci-solution.js

TheOdinProject/javascript-exercises#368 Update sumAll-solution.js swap algorithm to standard method

TheOdinProject/javascript-exercises#369 Reformat test parameters for consistency

PolicyEngine/policyengine-app#1283 Render p in markdown as <p> instead of <div> for blogs

PolicyEngine/policyengine-app#1269 Rename title in PolicyBreakdown component

fullstack-hy2020/fullstack-hy2020.github.io#2981 Update part3b.md (add note on fly.io deployment error)

fullstack-hy2020/fullstack-hy2020.github.io#2980 Update part9f.md (fix incorrect link)

fullstack-hy2020/fullstack-hy2020.github.io#2979 Update part3d.md (fix incorrect links)

A Guide on Overcoming Nerd Snipe, Info Junk, and FOMO