add files 5.3

人工智能工程师课程补充资料合集/GPT模型/后GPT4时代-什么是 Prompt.md

+## 什么是 Prompt
+## 什么是 Prompt
+
+## From a news
+
+<img width="793" alt="image" src="https://user-images.githubusercontent.com/14297658/230248003-c3d30c1c-bef1-4744-a0f8-82f2fc5067c5.png">
+
+## 什么是 Prompt
+
+人类说话的时候是见人下菜的，见人说人话，见鬼说鬼话，会结合当前的环境进行调整
+但是对于AI而言，她是没有没有环境的前提假设的，所以我们需要给AI一个环境来帮助他理解人类的处境
+
+- 没有上下文的问答
+
+<img width="782" alt="image" src="https://user-images.githubusercontent.com/14297658/230248219-4c51fd4e-93bf-4e2a-96fc-afd6c6f7aae2.png">
+
+- 包含了一定上下文的问答
+
+<img width="716" alt="image" src="https://user-images.githubusercontent.com/14297658/230248384-9eb650ca-2922-4510-ac10-790c4c57e415.png">
+
+## 如何设计 Prompt，以及怎么样才能得到更好的 Prompt
+
+<img width="744" alt="image" src="https://user-images.githubusercontent.com/14297658/230248675-59472e62-748b-4914-ba96-f705920bbb27.png">
+<img width="720" alt="image" src="https://user-images.githubusercontent.com/14297658/230251842-9df052e8-7fad-44b3-8574-e4829aa3b81d.png">
+<img width="726" alt="image" src="https://user-images.githubusercontent.com/14297658/230251928-3bb58e0a-c0c7-401f-98ac-d4b8e9ea688d.png">
+
+- 目标
+<img width="725" alt="image" src="https://user-images.githubusercontent.com/14297658/230249190-3ef5b4df-eec2-4fa9-b76e-311dfc2a0d3c.png">
+<img width="725" alt="image" src="https://user-images.githubusercontent.com/14297658/230249285-c53cdad1-6fa2-4a8e-91e5-f045d40a9ee9.png">
+
+- 语言
+
+<img width="719" alt="image" src="https://user-images.githubusercontent.com/14297658/230249453-ed0d1a0f-fe28-45db-ac1e-1725c11f13ba.png">
+
+- 上下文
+
+<img width="721" alt="image" src="https://user-images.githubusercontent.com/14297658/230249667-b7a83517-7f5a-4500-98b6-3e84ab23a92d.png">
+
+- 多样性
+
+<img width="732" alt="image" src="https://user-images.githubusercontent.com/14297658/230250290-01e0df33-007e-4264-a462-1661ec52fbe0.png">
+
+- 质量评估
+
+<img width="742" alt="image" src="https://user-images.githubusercontent.com/14297658/230250752-930d2700-38ea-4254-9d26-4fd579cfd967.png">
+<img width="679" alt="image" src="https://user-images.githubusercontent.com/14297658/230251295-fb1560fe-5775-424d-ba2e-e613e344d803.png">
+
+## Prompt before and after
+
+|           | 前ChatGPT时代 | 后GPT时代 |
+|-----------|----------------------|------------------|
+| 为什么之前很少有人提到prompt？ | 缺乏对语言模型的关注，大部分时候使用的是专用的小模型，小模型对 Prompt不敏感，训练代价低没人拿来做噱头 | 重新认识到语言模型的重要性，通用和跨领域达模型月来越多，大模型对prompt更加敏感，切代价更高 |
+| QA（问答）和Prompt的区别 | QA侧重回答特定问题，Prompt提供上下文和指导生成文本，QA中模型对任务的理解隐藏在模型参数中，不需要显性提出 | Prompt提供更广泛的上下文，可以引导文本生成超出回答特定问题的范围，大模型只训练通用任务，所以在遇到专用任务时，需要设计更好的提示信息 |
+| Prompt由什么组成？ | Prompt可以由关键词、短语或完整句子组成；任务描述，任务目标，和例子，不提供例子就是0-shot | Prompt还可以包括其他上下文，例如主题或任务信息，提供n个例子就是n-shot|
+| 如何改进Prompt？ | Hard/Soft Imporvement 找到与具体任务相关的token进行搜索和优化 | 使用机器学习自动生成高质量的prompt，询问 Chat GPT 如何生存高质量的Promopt |

人工智能工程师课程补充资料合集/GPT模型/备注.txt

+Prompt的 doc，推荐还是在gist上面看，地址如下
+Prompt的 doc，推荐还是在gist上面看，地址如下
+https://gist.github.com/sawyerbutton/2305cd60e134ffc06d8402fbfc4b7035
\ No newline at end of file

人工智能工程师课程补充资料合集/GoogLeNet论文精读/GoogleNet.py

+from re import S
+from re import S
+import torch.nn as nn
+import torch
+import torch.nn.functional as F
+
+from utils.CustomLayers import ConvActivation, Inception
+
+class Auxiliary_classcification(nn.Module):
+    def __init__(self, input_channels, num_classes):
+        super().__init__()
+        self.avgpool = nn.AvgPool2d(kernel_size=5, stride=3)
+        self.conv = ConvActivation(input_channels, 128, kernel_size=1) # output[batch, 128, 4, 4]
+        self.fc1 = nn.Linear(2048, 1024)
+        self.fc2 = nn.Linear(1024, num_classes)
+    def forward(self, x):
+        # aux1: N x 512 x 14 x 14, aux2: N x 528 x 14 x 14
+        x = self.avgpool(x)
+        # aux1: N x 512 x 4 x 4, aux2: N x 528 x 4 x 4
+        x = self.conv(x)
+        # N x 128 x 4 x 4
+        x = torch.flatten(x)
+        
+        x = F.dropout(x, p=0.5, training=self.training)
+        x = self.fc1(x)
+        x = F.relu(x, inplace=True)
+        x = F.dropout(x, p=0.5, training=self.training)
+        x = self.fc2(x)
+        return x
+
+class GoogLeNet(nn.Module):
+    def __init__(self, num_classes=None, aux_logits=False, init_weights=False):
+        super().__init__()
+        self.aux_logits = aux_logits
+        
+        self.conv1 = ConvActivation(3, 64, kernel_size=7, stride=2, padding=3)
+        self.maxpool = nn.MaxPool2d(3, stride=2, ceil_mode=True)
+        
+        self.conv2 = ConvActivation(64, 192, kernel_size=3, padding=1)
+        self.maxpool2 = nn.MaxPool2d(kernel_size=3, stride=2, ceil_mode=True)
+        
+        # input_channels, out_nc1x1, out_nc3x3_reduce, out_nc3x3, out_nc5x5_reduce, out_nc5x5, out_nc_pool
+        self.inception3a = Inception(192, 64, 96, 128, 16, 32, 32)
+        self.inception3b = Inception(256, 128, 128, 192, 32, 96, 64)
+        self.maxpool3 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
+
+        self.inception4a = Inception(480, 192, 96, 208, 16, 48, 64)
+        self.inception4b = Inception(512, 160, 112, 224, 24, 64, 64)
+        self.inception4c = Inception(512, 128, 128, 256, 24, 64, 64)
+        self.inception4d = Inception(512, 112, 144, 288, 32, 64, 64)
+        self.inception4e = Inception(528, 256, 160, 320, 32, 128, 128)
+        self.maxpool4 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
+
+        self.inception5a = Inception(832, 256, 160, 320, 32, 128, 128)
+        self.inception5b = Inception(832, 384, 192, 384, 48, 128, 128)
+
+        if self.aux_logits:
+            self.aux1 = Auxiliary_classcification(input_channels=512, num_classes=num_classes)
+            self.aux2 = Auxiliary_classcification(input_channels=528, num_classes=num_classes)  
+        
+        self.avgpool = nn.AdaptiveAvgPool2d((1,1))
+        self.dropout = nn.Dropout(0.4)
+        self.fc = nn.Linear(1024, num_classes)
+        
+        if init_weights:
+            self._initial_weights()          
+    
+    def forward(self, x):
+        x = self.conv1(x)           # [b*3*224*224]  --> [b*64*112*112]
+        x = self.maxpool(x)         # [b*64*112*112] --> [b*64*56*56]
+        x = self.conv2(x)           # [b*64*56*56]   --> [b*192*56*56]
+        x = self.maxpool2(x)        # [b*192*56*56]  --> [b*192*28*28]
+        
+        x = self.inception3a(x)     # [b*192*28*28]  --> [b*256*28*28]
+        x = self.inception3b(x)     # [b*256*28*28]  --> [b*480*28*28]
+        x = self.maxpool3(x)        # [b*256*14*14]  --> [b*480*14*14]
+        
+        x = self.inception4a(x)     # [b*480*14*14]  --> [b*512*14*14]
+        if self.training and self.aux_logits:  # eval model lose this layer
+            aux1 = self.aux1(x)       
+        x = self.inception4b(x)     # [b*512*14*14]  --> [b*512*14*14]
+        x = self.inception4c(x)     # [b*512*14*14]  --> [b*512*14*14]
+        x = self.inception4d(x)     # [b*512*14*14]  --> [b*528*14*14]
+        if self.training and self.aux_logits:   # eval model lose this layer
+            aux2 = self.aux2(x)
+        x = self.inception4e(x)     # [b*528*14*14]  --> [b*832*14*14]
+        x = self.maxpool3(x)        # [b*832*7*7]    --> [b*832*7*7]
+        
+        x = self.inception5a(x)     # [b*832*7*7]    --> [b*832*7*7]
+        x = self.inception5b(x)     # [b*832*7*7]    --> [b*1024*7*7]
+        
+        x = self.avgpool(x)         # [b*1027*7*7]   --> [b*1024*1*1]
+        x = torch.flatten(x, 1)     # [b*1027*1*1]   --> [b*1024]
+        x = self.fc(x)              # [b*1024]       --> [b*num_classes]
+        
+        if self.training and self.aux_logits:   # eval model lose this layer
+            return x, aux2, aux1
+        return x
+
+    def _initial_weights(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+                if m.bias is not None:
+                    nn.init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                nn.init.normal_(m.weight, 0, 0.01)
+                nn.init.constant_(m.bias, 0)
\ No newline at end of file