David Thorsley

microsoft/Phi-3-mini-4k-instruct fails with exceptions

2

**Describe the bug** Phi-3-mini-4k-instruct does not run with Automodel. I've tried two cases: 1. using `nemo_automodel.NeMoAutoModelForCausalLM.from_pretrained` as the model `_target_` in the config. In this case I get the error...

feat: deepseek-v2 backward compatability

3

Adds support for greedy expert selection to the MoE layers, which will allow the DeepSeek V3 model to run DeepSeek V2 checkpoints. Label r0.1.0 added to request a cherry-pick if...