fix loading chatglm3 (#1937)

* update welcome svg

* update welcome message

* fix loading chatglm3

---------

Co-authored-by: binary-husky <qingxu.fu@outlook.com>
Co-authored-by: binary-husky <96192199+binary-husky@users.noreply.github.com>

request_llms/bridge_chatglm3.py

@@ -18,7 +18,7 @@ class GetGLM3Handle(LocalLLMHandle):
 
     def load_model_and_tokenizer(self):
         # 🏃‍♂️🏃‍♂️🏃‍♂️ 子进程执行
-        from transformers import AutoModel, AutoTokenizer
+        from transformers import AutoModel, AutoTokenizer, BitsAndBytesConfig
         import os, glob
         import os
         import platform
@@ -45,15 +45,13 @@ class GetGLM3Handle(LocalLLMHandle):
                 chatglm_model = AutoModel.from_pretrained(
                     pretrained_model_name_or_path=_model_name_,
                     trust_remote_code=True,
-                    device="cuda",
+                    quantization_config=BitsAndBytesConfig(load_in_4bit=True),
-                    load_in_4bit=True,
                 )
             elif LOCAL_MODEL_QUANT == "INT8":  # INT8
                 chatglm_model = AutoModel.from_pretrained(
                     pretrained_model_name_or_path=_model_name_,
                     trust_remote_code=True,
-                    device="cuda",
+                    quantization_config=BitsAndBytesConfig(load_in_8bit=True),
-                    load_in_8bit=True,
                 )
             else:
                 chatglm_model = AutoModel.from_pretrained(

requirements.txt

@@ -7,7 +7,7 @@ tiktoken>=0.3.3
 requests[socks]
 pydantic==2.5.2
 protobuf==3.18
-transformers>=4.27.1
+transformers>=4.27.1,<4.42
 scipdf_parser>=0.52
 anthropic>=0.18.1
 python-markdown-math

themes/welcome.js

@@ -8,7 +8,7 @@ class WelcomeMessage {
                 url: "https://github.com/binary-husky/gpt_academic/wiki/%E9%A1%B9%E7%9B%AE%E9%85%8D%E7%BD%AE%E8%AF%B4%E6%98%8E",
             },
             {
-                title: "Arxiv论文一键翻译",
+                title: "Arxiv论文翻译",
                 content: "无缝切换学术阅读语言，最优英文转中文的学术论文阅读体验。",
                 svg: "file=themes/svg/arxiv.svg",
                 url: "https://www.bilibili.com/video/BV1dz4y1v77A/",
@@ -67,6 +67,18 @@ class WelcomeMessage {
                 svg: "file=themes/svg/check.svg",
                 url: "https://github.com/binary-husky/gpt_academic/wiki",
             },
+            {
+                title: "接入更多新模型",
+                content: "模型迭代日新月异，一起动手接入更多新的在线或本地大模型吧。",
+                svg: "file=themes/svg/box.svg",
+                url: "https://github.com/binary-husky/gpt_academic/blob/master/request_llms/README.md",
+            },
+            {
+                title: "联动VLLM等服务",
+                content: "借助VLLM和OneApi等第三方服务高效地部署和运行大模型。",
+                svg: "file=themes/svg/location.svg",
+                url: "https://github.com/binary-husky/gpt_academic/wiki/如何更便捷地接入one-api",
+            },
         ];
         this.visible = false;
         this.max_welcome_card_num = 6;