Yucheng Tu

A walkaround on JIT compilation issue in MNMG training

2

# Description > :memo: removing jit_compile=True, and annotate the whole train_step function by tf.function. > > * Please also include relevant motivation and context. When training a model with multiworkermirrored...

Faketensor issue when using torch inductor as backend with Trainer API

3

### System Info - `transformers` version: 4.27.0.dev0 - Platform: Linux-5.15.0-1030-aws-x86_64-with-glibc2.10 - Python version: 3.8.0 - Huggingface_hub version: 0.12.1 - PyTorch version (GPU?): 2.0.0a0+git45d775c (True) - Tensorflow version (GPU?): not installed...