有相关onnx模型转换与调用的支持嘛？

[Gmgge]

你好，非常感谢你们开源如此棒的工作

我在尝试将模型转换为onnx，以使用推理框架onnxruntime。
我进行了以下的工作
1.参考huggingface/transformers，调用模型转换将预训练模型转换为onnx，参考教程在此
运行脚本为：
python -m transformers.onnx --model "/project/gm/project/transformers/albert-chinese-tiny-train/" albert-chinese-tiny-train
结果为：
Some weights of the model checkpoint at /project/gm/project/transformers/albert-chinese-tiny-train/ were not used when initializing AlbertModel: ['predictions.dense.bias', 'predics.decoder.bias', 'predictions.LayerNorm.weight', 'predictions.LayerNorm.bias', 'predictions.decoder.weight', 'predictions.dense.weight']

• This IS expected if you are initializing AlbertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClasBertForPreTraining model).
• This IS NOT expected if you are initializing AlbertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification modeClassification model).
  Some weights of AlbertModel were not initialized from the model checkpoint at /project/gm/project/transformers/albert-chinese-tiny-train/ and are newly initialized: ['albert.pooler.bias']
  You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
  Using framework PyTorch: 1.8.1+cu102
  Validating ONNX model...
  -[✓] ONNX model output names match reference model ({'last_hidden_state'})
  - Validating ONNX Model output "last_hidden_state":
  -[✓] (2, 8, 312) matches (2, 8, 312)
  -[✓] all values close (atol: 1e-05)
  All good, model saved at: albert-chinese-tiny-train/model.onnx

2.参考transformers中对onnx的模型调用，结果提示参数错误
session = InferenceSession("onnx/model.onnx")
inputs = tokenizer("Using DistilBERT with ONNX Runtime!", return_tensors="np")
outputs = session.run(output_names=["last_hidden_state"], input_feed=dict(inputs))

有相关提示或者文档嘛？我阅读了本项目的文档，但是没有找到相关信息，不知道是否因为有些遗漏。

如果有相关的提示或者样例代码以参考，非常感谢