为各个模型添加了个性化软上限

ChuanhuChatbot.py

@@ -331,7 +331,7 @@ with gr.Blocks(css=customCSS, theme=small_and_beautiful_theme) as demo:
             token_count,
             top_p,
             temperature,
-            gr.State(max_token_streaming//2 if use_streaming_checkbox.value else max_token_all//2),
+            gr.State(sum(token_count.value[-4:])),
             model_select_dropdown,
             language_select_dropdown,
         ],

modules/chat_func.py

@@ -350,29 +350,29 @@ def predict(
             + colorama.Style.RESET_ALL
         )
 
-    # if stream:
-    #     max_token = max_token_streaming
-    # else:
-    #     max_token = max_token_all
-
-    # if sum(all_token_counts) > max_token and should_check_token_count:
-    #     status_text = f"精简token中{all_token_counts}/{max_token}"
-    #     logging.info(status_text)
-    #     yield chatbot, history, status_text, all_token_counts
-    #     iter = reduce_token_size(
-    #         openai_api_key,
-    #         system_prompt,
-    #         history,
-    #         chatbot,
-    #         all_token_counts,
-    #         top_p,
-    #         temperature,
-    #         max_token//2,
-    #         selected_model=selected_model,
-    #     )
-    #     for chatbot, history, status_text, all_token_counts in iter:
-    #         status_text = f"Token 达到上限，已自动降低Token计数至 {status_text}"
-    #         yield chatbot, history, status_text, all_token_counts
+    if stream:
+        max_token = MODEL_SOFT_TOKEN_LIMIT[selected_model]["streaming"]
+    else:
+        max_token = MODEL_SOFT_TOKEN_LIMIT[selected_model]["all"]
+
+    if sum(all_token_counts) > max_token and should_check_token_count:
+        status_text = f"精简token中{all_token_counts}/{max_token}"
+        logging.info(status_text)
+        yield chatbot, history, status_text, all_token_counts
+        iter = reduce_token_size(
+            openai_api_key,
+            system_prompt,
+            history,
+            chatbot,
+            all_token_counts,
+            top_p,
+            temperature,
+            max_token//2,
+            selected_model=selected_model,
+        )
+        for chatbot, history, status_text, all_token_counts in iter:
+            status_text = f"Token 达到上限，已自动降低Token计数至 {status_text}"
+            yield chatbot, history, status_text, all_token_counts
 
 
 def retry(

modules/presets.py

@@ -19,9 +19,7 @@ ssl_error_prompt = "SSL错误，无法获取对话。"  # SSL 错误
 no_apikey_msg = "API key长度不是51位，请检查是否输入正确。"  # API key 长度不足 51 位
 no_input_msg = "请输入对话内容。"  # 未输入对话内容
 
-max_token_streaming = 3500  # 流式对话时的最大 token 数
 timeout_streaming = 10  # 流式对话时的超时时间
-max_token_all = 3500  # 非流式对话时的最大 token 数
 timeout_all = 200  # 非流式对话时的超时时间
 enable_streaming_option = True  # 是否启用选择选择是否实时显示回答的勾选框
 HIDE_MY_KEY = False  # 如果你想在UI中隐藏你的 API 密钥，将此值设置为 True
@@ -57,6 +55,33 @@ MODELS = [
     "gpt-4-32k-0314",
 ]  # 可选的模型
 
+MODEL_SOFT_TOKEN_LIMIT = {
+    "gpt-3.5-turbo": {
+        "streaming": 3500,
+        "all": 3500
+    },
+    "gpt-3.5-turbo-0301": {
+        "streaming": 3500,
+        "all": 3500
+    },
+    "gpt-4": {
+        "streaming": 7500,
+        "all": 7500
+    },
+    "gpt-4-0314": {
+        "streaming": 7500,
+        "all": 7500
+    },
+    "gpt-4-32k": {
+        "streaming": 31000,
+        "all": 31000
+    },
+    "gpt-4-32k-0314": {
+        "streaming": 31000,
+        "all": 31000
+    }
+}
+
 REPLY_LANGUAGES = [
     "中文",
     "English",