from transformers import AutoConfig, AutoModel, AutoTokenizer

1.在线加载

model = AutoModel.from_pretrained("hfl/rbt3", force_download=True)

---

2.模型下载

!git clone "https://huggingface.co/hfl/rbt3"!git lfs clone "https://huggingface.co/hfl/rbt3" --include="*.bin"
# 过滤掉tf文件

---

3.离线加载

model = AutoModel.from_pretrained("rbt3")

---

4.模型加载参数

model = AutoModel.from_pretrained("rbt3")model.config
'''
BertConfig {"_name_or_path": "rbt3","architectures": ["BertForMaskedLM"],"attention_probs_dropout_prob": 0.1,"classifier_dropout": null,"directionality": "bidi","hidden_act": "gelu","hidden_dropout_prob": 0.1,"hidden_size": 768,"initializer_range": 0.02,"intermediate_size": 3072,"layer_norm_eps": 1e-12,"max_position_embeddings": 512,"model_type": "bert","num_attention_heads": 12,"num_hidden_layers": 3,"output_past": true,"pad_token_id": 0,"pooler_fc_size": 768,"pooler_num_attention_heads": 12,"pooler_num_fc_layers": 3,"pooler_size_per_head": 128,"pooler_type": "first_token_transform",
..."transformers_version": "4.28.1","type_vocab_size": 2,"use_cache": true,"vocab_size": 21128
}
'''

---

config = AutoConfig.from_pretrained("./rbt3/")

查看参数同pipeline，看config的基类代码，具体方法也是

from transformers import BertConfig

---

5.模型调用

sen = "弱小的我也有大梦想！"
tokenizer = AutoTokenizer.from_pretrained("rbt3")
inputs = tokenizer(sen, return_tensors="pt")
inputs
'''
{
'input_ids': tensor([[ 101, 2483, 2207, 4638, 2769,  738, 3300, 1920, 3457, 2682, 8013,  102]]), 
'token_type_ids': tensor([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]),
'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]])}
'''

---

5.1不带Model Head的模型调用

model = AutoModel.from_pretrained("rbt3", output_attentions=True)output = model(**inputs)
output
'''
BaseModelOutputWithPoolingAndCrossAttentions(last_hidden_state=tensor([[[ 0.6804,  0.6664,  0.7170,  ..., -0.4102,  0.7839, -0.0262],[-0.7378, -0.2748,  0.5034,  ..., -0.1359, -0.4331, -0.5874],[-0.0212,  0.5642,  0.1032,  ..., -0.3617,  0.4646, -0.4747],...,[ 0.0853,  0.6679, -0.1757,  ..., -0.0942,  0.4664,  0.2925],[ 0.3336,  0.3224, -0.3355,  ..., -0.3262,  0.2532, -0.2507],[ 0.6761,  0.6688,  0.7154,  ..., -0.4083,  0.7824, -0.0224]]],grad_fn=<NativeLayerNormBackward0>), pooler_output=tensor([[-1.2646e-01, -9.8619e-01, -1.0000e+00, -9.8325e-01,  8.0238e-01,-6.6268e-02,  6.6919e-02,  1.4784e-01,  9.9451e-01,  9.9995e-01,-8.3051e-02, -1.0000e+00, -9.8865e-02,  9.9980e-01, -1.0000e+00,9.9993e-01,  9.8291e-01,  9.5363e-01, -9.9948e-01, -1.3219e-01,-9.9733e-01, -7.7934e-01,  1.0720e-01,  9.8040e-01,  9.9953e-01,-9.9939e-01, -9.9997e-01,  1.4967e-01, -8.7627e-01, -9.9996e-01,-9.9821e-01, -9.9999e-01,  1.9396e-01, -1.1277e-01,  9.9359e-01,-9.9153e-01,  4.4752e-02, -9.8731e-01, -9.9942e-01, -9.9982e-01,2.9360e-02,  9.9847e-01, -9.2014e-03,  9.9999e-01,  1.7111e-01,4.5071e-03,  9.9998e-01,  9.9467e-01,  4.9726e-03, -9.0707e-01,6.9056e-02, -1.8141e-01, -9.8831e-01,  9.9668e-01,  4.9800e-01,1.2997e-01,  9.9895e-01, -1.0000e+00, -9.9990e-01,  9.9478e-01,-9.9989e-01,  9.9906e-01,  9.9820e-01,  9.9990e-01, -6.8953e-01,9.9990e-01,  9.9987e-01,  9.4563e-01, -3.7660e-01, -1.0000e+00,1.3151e-01, -9.7371e-01, -9.9997e-01, -1.3228e-02, -2.9801e-01,-9.9985e-01,  9.9662e-01, -2.0004e-01,  9.9997e-01,  3.6876e-01,-9.9997e-01,  1.5462e-01,  1.9265e-01,  8.9871e-02,  9.9996e-01,9.9998e-01,  1.5184e-01, -8.9714e-01, -2.1646e-01, -9.9922e-01,
...1.7911e-02, 4.8672e-01],[4.0732e-01, 3.8137e-02, 9.6832e-03,  ..., 4.4490e-02,2.2997e-02, 4.0793e-01],[1.7047e-01, 3.6989e-02, 2.3646e-02,  ..., 4.6833e-02,2.5233e-01, 1.6721e-01]]]], grad_fn=<SoftmaxBackward0>)), cross_attentions=None)
Output is truncated. View as a scrollable element or open in a text editor. Adjust cell output settings...
'''

---

output.last_hidden_state.size()
'''
torch.Size([1, 12, 768])
'''

---

5.2带Model Head的模型调用

from transformers import AutoModelForSequenceClassification, BertForSequenceClassificationclz_model = AutoModelForSequenceClassification.from_pretrained("rbt3", num_labels=10)
clz_model(**inputs)
'''
SequenceClassifierOutput(loss=None, logits=tensor([[-0.1776,  0.2208, -0.5060, -0.3938, -0.5837,  1.0171, -0.2616,  0.0495,0.1728,  0.3047]], grad_fn=<AddmmBackward0>), hidden_states=None, attentions=None)
'''clz_model.config.num_labels
# 2

---