Multinode Quantization

[Gran-0]

Does TensorRT-LLM support multinode quantization when checkpoint shards are stored on different machines? For example, using 2 nodes with 8 GPUs each to quantize a model with TP size 16. Using 2 nodes ssh and Openmpi, I'm encountering the error below.

Error:

NotImplementedError: ModelOpt does not support Dtensor sharded models to quantization/restore entry point yet. Please shard the model after quantization/restore