Commits

Branch selector

 main

User selector

odelalleau

Datepicker

All time

Commit History

Commits on Apr 22, 2024

• Skip top_p computations when set to 1.0 (NVIDIA#8905)

  Show description for e0b3fe5
  

  odelalleau

  and

  pablo-garay

  authored
  e0b3fe5

  Copy full SHA for e0b3fe5

Commits on Jan 18, 2024

• Fix learning rate schedule in Megatron models when max_steps is not set (NVIDIA#7518)

  Show description for 7449c67
  odelalleau

  authored
  7449c67

  Copy full SHA for 7449c67

Commits on Dec 12, 2023

• Fix crash when converting to mcore a model using rotary embeddings (NVIDIA#7998)

  Show description for 8eaa504
  

  odelalleau

  and

  cuichenx

  authored
  8eaa504

  Copy full SHA for 8eaa504

Commits on Sep 27, 2023

• Add the strategy argument to MegatronGPTModel.generate() (NVIDIA#7264)

  Show description for a376063
  odelalleau

  authored
  a376063

  Copy full SHA for a376063

Commits on Aug 28, 2023

• Bugfix and optimization in end_of_generation_condition() (NVIDIA#7267)

  Show description for 6861215
  odelalleau

  authored
  6861215

  Copy full SHA for 6861215

Commits on Aug 13, 2023

• Fix consumed_samples which was off by one batch (NVIDIA#7212)

  Show description for f4afdf0
  

  odelalleau

  and

  MaximumEntropy

  authored
  f4afdf0

  Copy full SHA for f4afdf0