Yaroslav Bulatov

tensorflow-community-wheels. Place to upload links to TensorFlow wheels

stuff. Stuff I uploaded to share online or to access from a different machine

kfac_pytorch. Python

memory_util. TensorFlow util for building memory usage timeline from LOG_MEMORY messages

chain_constant_memory. Example of backprop which uses constant memory

imperative. imperative programming in TensorFlow

memory_probe_ops. TensorFlow kernels for probing memory

notebooks. Jupyter Notebook

lbfgs. Port of l-BFGS from Torch to Tensorflow

max_align_bytes_op. TensorFlow op to return EIGEN_MAX_ALIGN_BYTES

whitening. Mathematica

tf_build. TensorFlow build on AWS

history-of-deep-learning. HTML

bflm. bflm

newton. Mathematica

einograd. Python

early-eager. Python

kaczmarz. Python

dominic_densenet. Python

cluster. train on AWS

m5-gemm. Fast fp32 GEMM on Apple M5 Max — port of percisely.xyz/gemm to Metal 4 (replaces the blocked simdgroup_async_copy intrinsic with a synchronous coalesced load; matches/beats MPS at 1k–8k)