Vadim Dabravolski

Sharing an implementation of working TF Config for MultiNodeMirroredStrategy below. This has been tested on SageMaker Deep Learning container with TensorFlow v2.8 (link to [dockerfile](https://github.com/aws/deep-learning-containers/blob/master/tensorflow/training/docker/2.8/py3/cu112/Dockerfile.gpu)). ```python def _build_tf_config(): hosts =...

See reported D2 issue: https://github.com/facebookresearch/detectron2/issues/1566

Was interested in the same questions, @usuyama. See excerpt from Megatron paper. Does look like MegatronHF will require some updates on HF side. 

I had the same issue. Anyone resolved it?

The following worked for me (adding "super" call in spider init): ``` def __init__(self, *a, **kw): super(NewsSpider, self).__init__(*a, **kw) ```

Hi, adding +1 to @AlpinDale. We are interested to experiment with MLP speculator, specifically, on latest Llama3.1 models. Excellent work overall @JRosenkranz !

@pietro-fragnito hi, was this issue resolved? Appreciate if you could share the solution.