jburtoft's picture
v1: V-JEPA2 ViT-L encoder NKI kernels for Trainium2 (4.35x, 44% MFU)
eadc767
Raw History Blame Contribute Delete
442 Bytes
V-JEPA2 encoder NKI kernels for AWS Trainium2
Copyright 2026 (kernels: res_layernorm.py, attention_block.py, mlp.py, rope_tables.py,
model glue and tooling -- original work, Apache-2.0)
kernels/attention_inner.py is a modified copy of the AWS Neuron SDK's nkilib
core/attention/attention_cte.py, Copyright Amazon.com, Inc. or its affiliates,
licensed under the Apache License, Version 2.0. See the file header for the
list of modifications.