David

Signed-off-by: David Mosallanezhad # What does this PR do ? Fixed Megatron export for updated Megatron models **Collection**: [Note which collection this PR will affect] # Changelog - Changed tokens_head...

fixes invalid_ids error on Megatron decode function

1

Signed-off-by: David Mosallanezhad # What does this PR do ? fixes invalid_ids error on Megatron decode function by changing invalid tokens to **Collection**: NLP # Changelog - small bug fix...

Megatron NMT Adapters

15

Signed-off-by: David Mosallanezhad # What does this PR do ? Megatron NMT Adapters **Collection**: [Note which collection this PR will affect] # Changelog - Add specific line by line info...

Signed-off-by: David Mosallanezhad # What does this PR do ? adding decoder memory for faster inference - MegatronNMT **Collection**: [Note which collection this PR will affect] # Changelog - Add...

Updated MegatronNMT properties to detect Float16Module

1

# What does this PR do ? - Updated MegatronNMT properties to detect Float16Module - Updated the log_softmax function to match the dtypes in megatron_export **Collection**: [Note which collection this...

Davidm/gpt sft dataset fix

6

# What does this PR do ? When using small prompts, the GPT_SFT_DATASET generated input prompts because of this logic: ``` if len(ids) < truncation_length: logging.warning(f'{key} is not long enough...

Hi, Please add the following data: https://github.com/JieYingWu/COVID-19_US_County-level_Summaries It contains aggregated machine-readable file counties.csv with demographic, socioeconomic, health care, and education data for each county in the 50 states and Washington...