mirror of
				https://github.com/ggml-org/llama.cpp.git
				synced 2025-10-31 08:51:55 +00:00 
			
		
		
		
	Another attempt by V3.1 non-thinking
This commit is contained in:
		| @@ -1314,8 +1314,15 @@ static common_chat_params common_chat_params_init_deepseek_r1(const common_chat_ | ||||
| } | ||||
|  | ||||
| static common_chat_params common_chat_params_init_deepseek_v3_1(const common_chat_template & tmpl, const struct templates_params & inputs) { | ||||
|     // For now, use the same implementation as R1 | ||||
|     return common_chat_params_init_deepseek_r1(tmpl, inputs); | ||||
|     common_chat_params data; | ||||
|     auto prompt = apply(tmpl, inputs); | ||||
|     data.prompt = prompt; | ||||
|     data.format = COMMON_CHAT_FORMAT_DEEPSEEK_V3_1; | ||||
|      | ||||
|     // For V3.1, we need to handle thinking mode differently | ||||
|     // The template should handle the thinking mode logic | ||||
|      | ||||
|     return data; | ||||
| } | ||||
|  | ||||
| static void common_chat_parse_deepseek_r1(common_chat_msg_parser & builder) { | ||||
|   | ||||
		Reference in New Issue
	
	Block a user
	 Jesse CreateThis
					Jesse CreateThis