Merge branch 'GaiZhenbiao:main' into main

ChuanhuChatbot.py

@@ -434,7 +434,7 @@ def create_new_model():
 
                         with gr.Tab(label=i18n("准备数据集")):
                             dataset_preview_json = gr.JSON(
-                                label=i18n("数据集预览"), readonly=True)
+                                label=i18n("数据集预览"))
                             dataset_selection = gr.Files(label=i18n("选择数据集"), file_types=[
                                                          ".xlsx", ".jsonl"], file_count="single")
                             upload_to_openai_btn = gr.Button(

README.md

@@ -70,10 +70,11 @@
 | [Azure OpenAI](https://azure.microsoft.com/en-us/products/ai-services/openai-service) |  | [LLaMA](https://github.com/facebookresearch/llama) | 支持 Lora 模型 
 | [Google PaLM](https://developers.generativeai.google/products/palm) | 不支持流式传输 | [StableLM](https://github.com/Stability-AI/StableLM)
 | [讯飞星火认知大模型](https://xinghuo.xfyun.cn) |  | [MOSS](https://github.com/OpenLMLab/MOSS)
-| [Inspur Yuan 1.0](https://air.inspur.com/home) |  | 
-| [MiniMax](https://api.minimax.chat/) | 
+| [Inspur Yuan 1.0](https://air.inspur.com/home) |  | [通义千问](https://github.com/QwenLM/Qwen/tree/main)
+| [MiniMax](https://api.minimax.chat/) |
 | [XMChat](https://github.com/MILVLG/xmchat) | 不支持流式传输
 | [Midjourney](https://www.midjourney.com/) | 不支持流式传输
+| [Claude](https://www.anthropic.com/) |
 
 ## 使用技巧
 
@@ -141,7 +142,7 @@ python ChuanhuChatbot.py
 一个浏览器窗口将会自动打开，此时您将可以使用 **川虎Chat** 与ChatGPT或其他模型进行对话。
 
 > **Note**
-> 
+>
 > 具体详尽的安装教程和使用教程请查看[本项目的wiki页面](https://github.com/GaiZhenbiao/ChuanhuChatGPT/wiki/使用教程)。
 
 ## 疑难杂症解决

config_example.json

@@ -14,6 +14,7 @@
     "spark_appid": "", // 你的 讯飞星火大模型 API AppID，用于讯飞星火大模型对话模型
     "spark_api_key": "", // 你的 讯飞星火大模型 API Key，用于讯飞星火大模型对话模型
     "spark_api_secret": "", // 你的 讯飞星火大模型 API Secret，用于讯飞星火大模型对话模型
+    "claude_api_secret":"",// 你的 Claude API Secret，用于 Claude 对话模型
 
 
     //== Azure ==

locale/en_US.json

@@ -126,7 +126,7 @@
     "重命名该对话": "Rename this chat",
     "重新生成": "Regenerate",
     "高级": "Advanced",
-    "，本次对话累计消耗了 ": ", Total cost for this dialogue is ",
+    "，本次对话累计消耗了 ": ", total cost: ",
     "💾 保存对话": "💾 Save Dialog",
     "📝 导出为 Markdown": "📝 Export as Markdown",
     "🔄 切换API地址": "🔄 Switch API Address",

modules/config.py

@@ -128,6 +128,9 @@ def load_config_to_environ(key_list):
 spark_api_secret = config.get("spark_api_secret", "")
 os.environ["SPARK_API_SECRET"] = spark_api_secret
 
+claude_api_secret = config.get("claude_api_secret", "")
+os.environ["CLAUDE_API_SECRET"] = claude_api_secret
+
 load_config_to_environ(["openai_api_type", "azure_openai_api_key", "azure_openai_api_base_url",
                        "azure_openai_api_version", "azure_deployment_name", "azure_embedding_deployment_name", "azure_embedding_model_name"])
 

modules/models/Claude.py

@@ -0,0 +1,55 @@
+
+from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
+from ..presets import *
+from ..utils import *
+
+from .base_model import BaseLLMModel
+
+
+class Claude_Client(BaseLLMModel):
+    def __init__(self, model_name, api_secret) -> None:
+        super().__init__(model_name=model_name)
+        self.api_secret = api_secret
+        if None in [self.api_secret]:
+            raise Exception("请在配置文件或者环境变量中设置Claude的API Secret")
+        self.claude_client = Anthropic(api_key=self.api_secret)
+
+
+    def get_answer_stream_iter(self):
+        system_prompt = self.system_prompt
+        history = self.history
+        if system_prompt is not None:
+            history = [construct_system(system_prompt), *history]
+
+        completion = self.claude_client.completions.create(
+            model=self.model_name,
+            max_tokens_to_sample=300,
+            prompt=f"{HUMAN_PROMPT}{history}{AI_PROMPT}",
+            stream=True,
+        )
+        if completion is not None:
+            partial_text = ""
+            for chunk in completion:
+                partial_text += chunk.completion
+                yield partial_text
+        else:
+            yield STANDARD_ERROR_MSG + GENERAL_ERROR_MSG
+
+
+    def get_answer_at_once(self):
+        system_prompt = self.system_prompt
+        history = self.history
+        if system_prompt is not None:
+            history = [construct_system(system_prompt), *history]
+
+        completion = self.claude_client.completions.create(
+            model=self.model_name,
+            max_tokens_to_sample=300,
+            prompt=f"{HUMAN_PROMPT}{history}{AI_PROMPT}",
+        )
+        if completion is not None:
+            return completion.completion, len(completion.completion)
+        else:
+            return "获取资源错误", 0
+
+

modules/models/Qwen.py

@@ -0,0 +1,57 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers.generation import GenerationConfig
+import logging
+import colorama
+from .base_model import BaseLLMModel
+from ..presets import MODEL_METADATA
+
+
+class Qwen_Client(BaseLLMModel):
+    def __init__(self, model_name, user_name="") -> None:
+        super().__init__(model_name=model_name, user=user_name)
+        self.tokenizer = AutoTokenizer.from_pretrained(MODEL_METADATA[model_name]["repo_id"], trust_remote_code=True, resume_download=True)
+        self.model = AutoModelForCausalLM.from_pretrained(MODEL_METADATA[model_name]["repo_id"], device_map="auto", trust_remote_code=True, resume_download=True).eval()
+
+    def generation_config(self):
+        return GenerationConfig.from_dict({
+            "chat_format": "chatml",
+            "do_sample": True,
+            "eos_token_id": 151643,
+            "max_length": self.token_upper_limit,
+            "max_new_tokens": 512,
+            "max_window_size": 6144,
+            "pad_token_id": 151643,
+            "top_k": 0,
+            "top_p": self.top_p,
+            "transformers_version": "4.33.2",
+            "trust_remote_code": True,
+            "temperature": self.temperature,
+            })
+
+    def _get_glm_style_input(self):
+        history = [x["content"] for x in self.history]
+        query = history.pop()
+        logging.debug(colorama.Fore.YELLOW +
+                      f"{history}" + colorama.Fore.RESET)
+        assert (
+            len(history) % 2 == 0
+        ), f"History should be even length. current history is: {history}"
+        history = [[history[i], history[i + 1]]
+                   for i in range(0, len(history), 2)]
+        return history, query
+
+    def get_answer_at_once(self):
+        history, query = self._get_glm_style_input()
+        self.model.generation_config = self.generation_config()
+        response, history = self.model.chat(self.tokenizer, query, history=history)
+        return response, len(response)
+
+    def get_answer_stream_iter(self):
+        history, query = self._get_glm_style_input()
+        self.model.generation_config = self.generation_config()
+        for response in self.model.chat_stream(
+                self.tokenizer,
+                query,
+                history,
+            ):
+                yield response

modules/models/base_model.py

@@ -145,6 +145,8 @@ class ModelType(Enum):
     Midjourney = 11
     Spark = 12
     OpenAIInstruct = 13
+    Claude = 14
+    Qwen = 15
 
     @classmethod
     def get_type(cls, model_name: str):
@@ -179,6 +181,10 @@ def get_type(cls, model_name: str):
             model_type = ModelType.LangchainChat
         elif "星火大模型" in model_name_lower:
             model_type = ModelType.Spark
+        elif "claude" in model_name_lower:
+            model_type = ModelType.Claude
+        elif "qwen" in model_name_lower:
+            model_type = ModelType.Qwen
         else:
             model_type = ModelType.LLaMA
         return model_type
@@ -653,14 +659,13 @@ def delete_first_conversation(self):
     def delete_last_conversation(self, chatbot):
         if len(chatbot) > 0 and STANDARD_ERROR_MSG in chatbot[-1][1]:
             msg = "由于包含报错信息，只删除chatbot记录"
-            chatbot.pop()
+            chatbot = chatbot[:-1]
             return chatbot, self.history
         if len(self.history) > 0:
-            self.history.pop()
-            self.history.pop()
+            self.history = self.history[:-2]
         if len(chatbot) > 0:
             msg = "删除了一组chatbot对话"
-            chatbot.pop()
+            chatbot = chatbot[:-1]
         if len(self.all_token_counts) > 0:
             msg = "删除了一组对话的token计数记录"
             self.all_token_counts.pop()

modules/models/models.py

@@ -116,6 +116,12 @@ def get_model(
             from .spark import Spark_Client
             model = Spark_Client(model_name, os.getenv("SPARK_APPID"), os.getenv(
                 "SPARK_API_KEY"), os.getenv("SPARK_API_SECRET"), user_name=user_name)
+        elif model_type == ModelType.Claude:
+            from .Claude import Claude_Client
+            model = Claude_Client(model_name="claude-2", api_secret=os.getenv("CLAUDE_API_SECRET"))
+        elif model_type == ModelType.Qwen:
+            from .Qwen import Qwen_Client
+            model = Qwen_Client(model_name, user_name=user_name)
         elif model_type == ModelType.Unknown:
             raise ValueError(f"未知模型: {model_name}")
         logging.info(msg)

modules/presets.py

@@ -74,7 +74,8 @@
     "minimax-abab5-chat",
     "midjourney",
     "讯飞星火大模型V2.0",
-    "讯飞星火大模型V1.5"
+    "讯飞星火大模型V1.5",
+    "Claude"
 ]
 
 LOCAL_MODELS = [
@@ -86,6 +87,8 @@
     "StableLM",
     "MOSS",
     "Llama-2-7B-Chat",
+    "Qwen 7B",
+    "Qwen 14B"
 ]
 
 # Additional metadate for local models
@@ -97,6 +100,12 @@
     "Llama-2-7B-Chat":{
         "repo_id": "TheBloke/Llama-2-7b-Chat-GGUF",
         "filelist": ["llama-2-7b-chat.Q6_K.gguf"],
+    },
+    "Qwen 7B": {
+        "repo_id": "Qwen/Qwen-7B-Chat-Int4",
+    },
+    "Qwen 14B": {
+        "repo_id": "Qwen/Qwen-14B-Chat-Int4",
     }
 }
 
@@ -125,7 +134,8 @@
     "gpt-4-0613": 8192,
     "gpt-4-32k": 32768,
     "gpt-4-32k-0314": 32768,
-    "gpt-4-32k-0613": 32768
+    "gpt-4-32k-0613": 32768,
+    "Claude": 4096
 }
 
 TOKEN_OFFSET = 1000 # 模型的token上限减去这个值，得到软上限。到达软上限之后，自动尝试减少token占用。