How to use test/ls_quant_bert.py?

[gjm441]

When I run the scripts, first meet a error: not has model.bin? Then, I download bert-base-uncased model (pytorch_model.bin)from huggingface: git lfs install,git clone https://huggingface.co/bert-base-uncased.
I use pytorch_model.bin as ckpt_path(model_name), output errors:

bert.encoder.layer.10.self_attn.out_proj.output_quant: No amax in state_dict. bert.encoder.layer.10.self_attn.out_proj.weight_quant: No amax in state_dict. bert.encoder.layer.10.fc1.input_quant: No amax in state_dict. bert.encoder.layer.10.fc1.output_quant: No amax in state_dict. bert.encoder.layer.10.fc1.weight_quant: No amax in state_dict. bert.encoder.layer.10.fc2.input_quant: No amax in state_dict. bert.encoder.layer.10.fc2.output_quant: No amax in state_dict. bert.encoder.layer.10.fc2.weight_quant: No amax in state_dict. bert.encoder.layer.11.self_attn.qkv_proj.input_quant: No amax in state_dict. bert.encoder.layer.11.self_attn.qkv_proj.output_quant: No amax in state_dict. bert.encoder.layer.11.self_attn.qkv_proj.weight_quant: No amax in state_dict. bert.encoder.layer.11.self_attn.out_proj.input_quant: No amax in state_dict. bert.encoder.layer.11.self_attn.out_proj.output_quant: No amax in state_dict. bert.encoder.layer.11.self_attn.out_proj.weight_quant: No amax in state_dict. bert.encoder.layer.11.fc1.input_quant: No amax in state_dict. bert.encoder.layer.11.fc1.output_quant: No amax in state_dict. bert.encoder.layer.11.fc1.weight_quant: No amax in state_dict. bert.encoder.layer.11.fc2.input_quant: No amax in state_dict. bert.encoder.layer.11.fc2.output_quant: No amax in state_dict. bert.encoder.layer.11.fc2.weight_quant: No amax in state_dict. creating lightseq model... Traceback (most recent call last): File "test/ls_quant_bert.py", line 204, in main() File "test/ls_quant_bert.py", line 181, in main ls_model = LightseqBertClassification(args.model, hf_model) File "test/ls_quant_bert.py", line 89, in init self.ls_bert = lsi.QuantBert(ls_weight_path, 8) RuntimeError: Unsupported weight extention for [pytorch_model.pt]; Supported extensions: .pb, .hdf5

[hexisyztem]

Need to convert the data format to the format of lightseq