Sudhakar Singh

icon indicating a website link linkedin.com/in/sudhakarsingh27/ icon indicating an email [email protected]

icon company @nvidia icon location San Jose, CA icon location LLMs, Parallel Systems, Quantum Computing

Results 70 comments of Sudhakar Singh

[BUG] context manager syntax bug in transformer_block.py

Acked!

fix context manager syntax bug in transformer_block.py

Thanks for spotting the bug! Working on fixing for the next release

Fp8 model init factory

The CM API needs to be then managed and added by the user. But this change would allow that in addition to just an argument. When trying to integrate with...

Expose `rotary_base` as an arg instead of hardcoding

/te-ci pytorch

feat: Apply RoPE embedding with sequence offsets

/ok to test

feat: Apply RoPE embedding with sequence offsets

/ok to test https://github.com/NVIDIA/Megatron-LM/pull/2416/commits/f0461ab06bf97267d25abf30b73686e31d3969f0

Enable SWA with CP for THD input format

/te-ci pytorch L0

Enable SWA with CP for THD input format

/te-ci pytorch L0

Enable SWA with CP for THD input format

/te-ci pytorch L1

Enable SWA with CP for THD input format

/te-ci pytorch