Open-source ML framework and AI hardware company. The tinygrad framework reduces all neural network ops to just three primitive types (elementwise, reduce, movement), making it unusually easy to port to new accelerators. The tinybox, their GPU workstation, benchmarked against MLPerf Training 4.0 systems costing 10x more.
Tiny Corp halves the price of its container-sized AI supercomputer to $5M by swapping speculative RDNA5 silicon for 256 shipping AMD MI300X accelerators.
Tiny Corp's 25,000-line open-source framework lands on the official MLPerf Training v6.0 leaderboard, training Llama 8B on AMD MI350X hardware with a fully custom software stack