This guide demonstrates the usage of the ONNX export functionality for Megatron models.
Set up the development environment by launching the latest NeMo container
The minimum version requirements for NeMo and TransformerEngine are below
nemo > 1.19
transformer_engine > 0.10
The export script supports the ONNX export of models with .nemo and .ckpt file extensions. The script also supports the export of the following types of models: GPT, T5, BERT, BART, NMT, RETRO. Commands for both file formats are discussed in the following sections. The model type used for the examples is GPT.
Export using .nemo file
A model with .nemo file extension can be exported using the command below
python3 examples/nlp/language_modeling/megatron_export.py \
model_type=gpt \
onnx_model_file=gpt_126m.onnx \
gpt_model_file=gpt_126m.nemo
Export using .ckpt file
A model with .ckpt file extension can be exported using the command below
python3 examples/nlp/language_modeling/megatron_export.py \
model_type=gpt \
onnx_model_file=gpt_126m.onnx \
checkpoint_dir=./gpt_126m/ \
checkpoint_name=model_weights.ckpt \
hparams_file=./gpt_126m/hparams.yaml