Adding Blob

app.py

@@ -1,7 +1,67 @@
 import gradio as gr
+import torch
+from transformers import BertModel
+from transformers import BertTokenizer
+import torch.nn as nn
 
-def greet(name):
-    return "Hello " + name + "!!"
+class BertSST2Model(nn.Module):
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
-iface.launch()
+    # 初始化类
+    def __init__(self, class_size, pretrained_name='bert-base-chinese'):
+        """
+        Args: 
+            class_size  :指定分类模型的最终类别数目，以确定线性分类器的映射维度
+            pretrained_name :用以指定bert的预训练模型
+        """
+        # 类继承的初始化，固定写法
+        super(BertSST2Model, self).__init__()
+        # 加载HuggingFace的BertModel
+        # BertModel的最终输出维度默认为768
+        # return_dict=True 可以使BertModel的输出具有dict属性，即以 bert_output['last_hidden_state'] 方式调用
+        self.bert = BertModel.from_pretrained(pretrained_name,
+                                              return_dict=True)
+        # 通过一个线性层将[CLS]标签对应的维度：768->class_size
+        # class_size 在SST-2情感分类任务中设置为：2
+        self.classifier = nn.Linear(768, class_size)
+
+    def forward(self, inputs):
+        # 获取DataLoader中已经处理好的输入数据：
+        # input_ids :tensor类型，shape=batch_size*max_len   max_len为当前batch中的最大句长
+        # input_tyi :tensor类型，
+        # input_attn_mask :tensor类型，因为input_ids中存在大量[Pad]填充，attention mask将pad部分值置为0，让模型只关注非pad部分
+        input_ids, input_tyi, input_attn_mask = inputs['input_ids'], inputs[
+            'token_type_ids'], inputs['attention_mask']
+        # 将三者输入进模型，如果想知道模型内部如何运作，前面的蛆以后再来探索吧~
+        output = self.bert(input_ids, input_tyi, input_attn_mask)
+        # bert_output 分为两个部分：
+        #   last_hidden_state:最后一个隐层的值
+        #   pooler output:对应的是[CLS]的输出,用于分类任务
+        # 通过线性层将维度：768->2
+        # categories_numberic：tensor类型，shape=batch_size*class_size，用于后续的CrossEntropy计算
+        categories_numberic = self.classifier(output.pooler_output)
+        return categories_numberic
+
+
+device = torch.device("cpu")
+pretrained_model_name = './bert-base-uncased'
+# 创建模型 BertSST2Model
+model = BertSST2Model(2, pretrained_model_name)
+# 固定写法，将模型加载到device上，
+# 如果是GPU上运行，此时可以观察到GPU的显存增加
+model.to(device)
+# 加载预训练模型对应的tokenizer
+tokenizer = BertTokenizer.from_pretrained(pretrained_model_name)
+
+
+def modelscope_quickstart(sentence):
+    inputs = tokenizer(sentence,
+                    padding=True,
+                    truncation=True,
+                    return_tensors="pt",
+                    max_length=512)
+    output = model(inputs)
+    cate = output.argmax(dim=1)
+    return "分类结果为：" + str(0 if cate else 1)
+
+demo = gr.Interface(fn=modelscope_quickstart, inputs="text", outputs="text")
+demo.launch()

bert-base-uncased/bert_config.json

@@ -0,0 +1,13 @@
+{
+  "attention_probs_dropout_prob": 0.1,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "max_position_embeddings": 512,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "type_vocab_size": 2,
+  "vocab_size": 30522
+}

bert-base-uncased/config.json

@@ -0,0 +1,23 @@
+{
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "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": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.6.0.dev0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

bert-base-uncased/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:097417381d6c7230bd9e3557456d726de6e83245ec8b24f529f60198a67b203a
+size 440473133

requirments.txt

@@ -0,0 +1,3 @@
+torch==1.8.0
+transformers==4.12.0
+tqdm