File tensor_parallel.py

examples/tensor_parallel.py:None–None · view source on GitHub ↗

Source from the content-addressed store, hash-verified

1	"""Tensor-parallel sampling across 2 GPUs.
2
3	Each GPU holds a shard of the vocabulary weights (V/2 rows each).
4	The fused kernel computes local Gumbel-max values, then an all-gather

nothing calls this directly

mainFunction · 0.70

no test coverage detected