From bed03a6ff11a2fd1c9981540c879bd1df01e4b85 Mon Sep 17 00:00:00 2001
From: myml <wurongjie1@gmail.com>
Date: Wed, 5 Apr 2023 01:05:06 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8Dchatglm=E6=A8=A1?=
 =?UTF-8?q?=E5=9E=8B=E8=A2=AB=E5=A4=8D=E5=88=B6=EF=BC=8C=E6=98=BE=E5=AD=98?=
 =?UTF-8?q?=E5=8D=A0=E7=94=A8=E8=BF=87=E5=A4=9A?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

model作为类成员会在类实例化时进行一次复制
这导致每询问一个问题显存占用就会翻倍
通过将model改成全局变量修复这个问题
---
 chatglm_llm.py | 18 ++++--------------
 1 file changed, 4 insertions(+), 14 deletions(-)

diff --git a/chatglm_llm.py b/chatglm_llm.py
index 8c00685..4a8b36a 100644
--- a/chatglm_llm.py
+++ b/chatglm_llm.py
@@ -5,24 +5,14 @@ from transformers import AutoTokenizer, AutoModel
 
 """ChatGLM_G is a wrapper around the ChatGLM model to fit LangChain framework. May not be an optimal implementation"""
 
+tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True)
+model = AutoModel.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True).half().cuda()
 
 class ChatGLM(LLM):
     max_token: int = 10000
     temperature: float = 0.1
     top_p = 0.9
     history = []
-    tokenizer = AutoTokenizer.from_pretrained(
-        "THUDM/chatglm-6b",
-        trust_remote_code=True
-    )
-    model = (
-        AutoModel.from_pretrained(
-            "THUDM/chatglm-6b",
-            trust_remote_code=True)
-        .half()
-        .cuda()
-    )
-
 
     def __init__(self):
         super().__init__()
@@ -34,8 +24,8 @@ class ChatGLM(LLM):
     def _call(self,
               prompt: str,
               stop: Optional[List[str]] = None) -> str:
-        response, updated_history = self.model.chat(
-            self.tokenizer,
+        response, updated_history = model.chat(
+            tokenizer,
             prompt,
             history=self.history,
             max_length=self.max_token,