Utils about Jamba model.
Make a dense model based on Jamba-v0.1 weights without using MoE:
python dense_downcycling.py --model_path ai21labs/Jamba-v0.1 --output_path test --expert_ids 0
Convert Jamba-v0.1 weights:
python convert_jamba_weights_to_hf.py --model_path ai21labs/Jamba-v0.1 --output_path test