支持gpt-4-v处理多张图片
This commit is contained in:
		
							parent
							
								
									9a21e13d33
								
							
						
					
					
						commit
						682898a3ba
					
				@ -198,7 +198,7 @@ def predict(inputs, llm_kwargs, plugin_kwargs, chatbot, history=[], system_promp
 | 
			
		||||
                    # 前者是API2D的结束条件,后者是OPENAI的结束条件
 | 
			
		||||
                    if ('data: [DONE]' in chunk_decoded) or (len(chunkjson['choices'][0]["delta"]) == 0):
 | 
			
		||||
                        # 判定为数据流的结束,gpt_replying_buffer也写完了
 | 
			
		||||
                        lastmsg = chatbot[-1][-1] + f"<br/><br/>{llm_kwargs['llm_model']}调用结束,该模型不具备上下文对话能力,如需追问,请及时切换模型。"
 | 
			
		||||
                        lastmsg = chatbot[-1][-1] + f"\n\n\n\n「{llm_kwargs['llm_model']}调用结束,该模型不具备上下文对话能力,如需追问,请及时切换模型。」"
 | 
			
		||||
                        yield from update_ui_lastest_msg(lastmsg, chatbot, history, delay=1)
 | 
			
		||||
                        logging.info(f'[response] {gpt_replying_buffer}')
 | 
			
		||||
                        break
 | 
			
		||||
 | 
			
		||||
		Loading…
	
	
			
			x
			
			
		
	
		Reference in New Issue
	
	Block a user