update

app.py

@@ -4,12 +4,11 @@ from langchain_core.messages import HumanMessage, AIMessage
 from llm import DeepSeekLLM
 from config import settings
 
+
 deep_seek_llm = DeepSeekLLM(api_key=settings.deep_seek_api_key)
-chat = ChatOpenAI(model=deep_seek_llm.default_model, api_key=deep_seek_llm.api_key, base_url=deep_seek_llm.base_url)
 
 
-def predict(message, history, model: str, temperature: float, max_tokens: int):
-    print('???model', model, temperature, max_tokens)
+def predict(message, history, chat):
     history_messages = []
     for human, assistant in history:
         history_messages.append(HumanMessage(content=human))
@@ -22,38 +21,52 @@ def predict(message, history, model: str, temperature: float, max_tokens: int):
         yield response_message
 
 
+def update_chat(_chat, _model: str, _temperature: float, _max_tokens: int):
+    _chat = ChatOpenAI(
+        model=_model,
+        api_key=deep_seek_llm.api_key,
+        base_url=deep_seek_llm.base_url,
+        temperature=_temperature,
+        max_tokens=_max_tokens,
+    )
+    return _chat
+
+
 with gr.Blocks() as app:
     with gr.Tab('聊天'):
+        chat_engine = gr.State(
+            value=ChatOpenAI(
+                model=deep_seek_llm.default_model,
+                api_key=deep_seek_llm.api_key,
+                base_url=deep_seek_llm.base_url,
+            )
+        )
         with gr.Row():
             with gr.Column(scale=2, min_width=600):
                 chatbot = gr.ChatInterface(
                     predict,
                     multimodal=True,
                     chatbot=gr.Chatbot(elem_id="chatbot", height=600),
-                    textbox=gr.MultimodalTextbox(),
-                    additional_inputs=[
-                        gr.Dropdown(choices=deep_seek_llm.support_models, label='模型'),
-                        gr.Slider(
+                    textbox=gr.MultimodalTextbox(lines=1),
+                    additional_inputs=[chat_engine]
+                )
+            with gr.Column(scale=1, min_width=300):
+                with gr.Accordion('Select Model', open=True):
+                    with gr.Column():
+                        model = gr.Dropdown(
+                            label='模型',
+                            choices=deep_seek_llm.support_models,
+                            value=deep_seek_llm.default_model
+                        )
+                        temperature = gr.Slider(
                             minimum=0.0,
                             maximum=1.0,
                             step=0.1,
+                            value=0.5,
                             label="Temperature",
                             key="temperature",
-                        ),
-                        gr.Number(
-                            minimum=1024,
-                            maximum=1024 * 20,
-                            step=128,
-                            value=4096,
-                            label="Max Tokens",
-                            key="max_tokens",
                         )
-                    ],
-                )
-            with gr.Column(scale=1, min_width=300):
-                with gr.Accordion('Select Model', open=True):
-                    with gr.Column():
-                        gr.Number(
+                        max_tokens = gr.Number(
                             minimum=1024,
                             maximum=1024 * 20,
                             step=128,
@@ -61,6 +74,9 @@ with gr.Blocks() as app:
                             label="Max Tokens",
                             key="max_tokens",
                         )
+        model.change(fn=update_chat, inputs=[chat_engine, model, temperature, max_tokens], outputs=[chat_engine])
+        temperature.change(fn=update_chat, inputs=[chat_engine, model, temperature, max_tokens], outputs=[chat_engine])
+        max_tokens.change(fn=update_chat, inputs=[chat_engine, model, temperature, max_tokens], outputs=[chat_engine])
 
     with gr.Tab('画图'):
         with gr.Row():

requirements-dev.lock

@@ -46,10 +46,6 @@ click==8.1.7
     # via nltk
     # via typer
     # via uvicorn
-colorama==0.4.6
-    # via click
-    # via tqdm
-    # via uvicorn
 contourpy==1.2.1
     # via matplotlib
 cycler==0.12.1
@@ -404,6 +400,8 @@ urllib3==2.2.2
 uvicorn==0.30.1
     # via fastapi
     # via gradio
+uvloop==0.19.0
+    # via uvicorn
 watchfiles==0.22.0
     # via uvicorn
 websockets==11.0.3

requirements.lock

@@ -46,10 +46,6 @@ click==8.1.7
     # via nltk
     # via typer
     # via uvicorn
-colorama==0.4.6
-    # via click
-    # via tqdm
-    # via uvicorn
 contourpy==1.2.1
     # via matplotlib
 cycler==0.12.1
@@ -404,6 +400,8 @@ urllib3==2.2.2
 uvicorn==0.30.1
     # via fastapi
     # via gradio
+uvloop==0.19.0
+    # via uvicorn
 watchfiles==0.22.0
     # via uvicorn
 websockets==11.0.3