✨ feature: add short-term-memory

.gitignore

@@ -16,7 +16,6 @@ output.txt
 *.mp4
 output/
 
-config/global.toml
 
 
 # converter

README.md

@@ -83,7 +83,7 @@ promopt = "./prompts/paimeng.txt"
 
 ## RoadMap
 
-- [ ] 支持对话的上下文
+- [x] 支持对话的上下文
 - [x] 支持自动识别录音开始和结束.(参考 Digtal_Life_Server)
 - [x] 支持流式切分长段落并且为句子.
 - [ ] 回答完成后可以保持激活一段时间
@@ -96,7 +96,7 @@ promopt = "./prompts/paimeng.txt"
 - [x] 接入 live2d 模型动画播放
 - [ ] 接入 mcp 情绪识别或者 BERT 情绪识别以及情绪动画播放
 - [ ] 接入网页端的对话框显示和支持
-- [ ] 优化 api-key 的调用, 用 streamlit 写一个配置文件的界面.
+- [x] 优化 api-key 的调用, 用 streamlit 写一个配置文件的界面.
 
 ## 原则
 

prompts/paimeng.txt

@@ -18,17 +18,10 @@
 “原来还有这种办法，我们快试试看！”
 “呜呜，虽然好生气，但不知道怎么反驳”
 “进不去！怎么想都进不去吧？”
-“前面的区域，以后再来探索吧！”
 “喂！现在怎么办？”
 “这样啊…看来我们只能明天再去找她了。”
 “欸？不行不行，怎么又是这种要求？”
-“呜哇，好丑的字！”
-“唔！史莱姆已经开始出现了吗，趁被发现前赶快回去酒馆吧！”
-“太好了，总算是来了个能拍板的人！”
-“哼哼，天外有天，人外有人！你在这里等着，我们马上就带他过来！”
-“呼…没想到这里居然会有毒蝎…”
 “对啊，很有道理！”
-“真是的，出门度个假还遇到这些家伙…”
 “一群不识相的家伙，哼！”
 “别发呆啦，你又看到了什么吗？”
 

src/chatbot/_dictionary.py

@@ -6,11 +6,14 @@
     from chatbot._typing import StSessionSateKeys
 
 session_keys: StSessionSateKeys = {
+    # for values
     "text_response": "text_response",
     "sentences": "sentences",
     "static_que": "static_que",
     "sentence_que": "sentence_que",
     "tts_que": "tts_que",
+    "short_term_memory": "short_term_memory",
+    # for settings.
     "initial_settings": "initial_settings",
     "sdk_base_url": "sdk_base_url",
     "sdk_key": "sdk_key",

src/chatbot/_typing.py

@@ -13,6 +13,7 @@ class StSessionSateKeys(TypedDict):
     static_que: str  # 静态队列，用于存储音频文件路径, 为了保持逻辑一致性设置
     sentence_que: str  # 动态队列，用于存储句子
     tts_que: str  # 动态队列，用于存储 TTS 音频文件路径
+    short_term_memory: str  # 记录单次问答的所有内容， 以 role: user , rool: asistant 进行记录。 list[dict[str,str]]
 
     # 配置文件
     initial_settings: str  # 初始设置，用于存储全局配置

src/chatbot/api/async_api.py

@@ -26,6 +26,10 @@
 if session_keys["text_response"] not in st.session_state:
     st.session_state[session_keys["text_response"]] = ""  # 初始化会话状态
 
+if session_keys["short_term_memory"] not in st.session_state:
+    st.session_state[session_keys["short_term_memory"]] = []  # 初始化短期记忆
+
+
 # --- Configuration ---
 settings = load_settings_file("config.toml", ServiceSettings)
 OPENAI_API_KEY = settings.sdk_key
@@ -38,7 +42,7 @@ async def get_openai_response_stream(
     prompt: str,
     model: str = MODEL,
     max_tokens: int = 15000,
-    temperature: float = 0.9,
+    temperature: float = 0.4,
     n: int = 1,
     stop: list[str] | None = None,
     presence_penalty: float = 0,
@@ -54,12 +58,13 @@ async def get_openai_response_stream(
         "Authorization": f"Bearer {OPENAI_API_KEY}",
         "Content-Type": "application/json",
     }
+    if len(st.session_state[session_keys["short_term_memory"]]) == 0:
+        # 如果短期记忆为空，添加系统提示
+        st.session_state[session_keys["short_term_memory"]].append({"role": "system", "content": SYSTEMPROMOT})
+    st.session_state[session_keys["short_term_memory"]].append({"role": "user", "content": prompt})
     data = {
         "model": model,
-        "messages": [
-            {"role": "system", "content": SYSTEMPROMOT},
-            {"role": "user", "content": prompt},
-        ],
+        "messages": st.session_state[session_keys["short_term_memory"]],
         "max_tokens": max_tokens,
         "temperature": temperature,
         "n": n,
@@ -91,11 +96,12 @@ async def get_openai_response_stream(
                         while True:
                             m = re.search(r"[。！？!?\.]", buffer)
                             if m:
-                                sentence = buffer[: m.end()]
+                                sentence = buffer[: m.end()].strip().replace("\n", "")
                                 # 第一次分句，记录耗时
                                 if first_sentence_time is None:
                                     first_sentence_time = time.monotonic()
                                     Logger.debug(f"首句耗时: {first_sentence_time - t_start:.3f} 秒")
+                                st.session_state[session_keys["text_response"]] += sentence
                                 yield sentence
                                 buffer = buffer[m.end() :]
                             else:
@@ -106,6 +112,10 @@ async def get_openai_response_stream(
     # 可选，总耗时打印
     t_end = time.monotonic()
     Logger.debug(f"openai 总耗时: {t_end - t_start:.3f} 秒")
+    st.session_state[session_keys["short_term_memory"]].append(
+        {"role": "assistant", "content": st.session_state[session_keys["text_response"]]}
+    )
+    Logger.debug(f"短期记忆:{st.session_state[session_keys['short_term_memory']]}")
 
 
 async def async_get_tts_response(text: str):

src/chatbot/api/sync_api.py

@@ -19,6 +19,8 @@
 
 if session_keys["text_response"] not in st.session_state:
     st.session_state[session_keys["text_response"]] = ""  # 初始化会话状态
+if session_keys["short_term_memory"] not in st.session_state:
+    st.session_state[session_keys["short_term_memory"]] = []  # 初始化短期记忆
 
 # --- Configuration ---
 settings = load_settings_file("config.toml", ServiceSettings)
@@ -32,7 +34,7 @@ def get_openai_response(
     prompt: str,
     model: str = MODEL,
     max_tokens: int = 15000,
-    temperature: float = 0.9,
+    temperature: float = 0.5,
     n: int = 1,
     stop: list[str] | None = None,
     presence_penalty: float = 0,
@@ -48,6 +50,11 @@ def get_openai_response(
         "Authorization": f"Bearer {OPENAI_API_KEY}",
         "Content-Type": "application/json",
     }
+    if len(st.session_state[session_keys["short_term_memory"]]) == 0:
+        st.session_state[session_keys["short_term_memory"]].append({"role": "system", "content": SYSTEMPROMOT})
+    st.session_state[session_keys["short_term_memory"]].append(
+        {"role": "user", "content": prompt}
+    )  # 添加用户输入到短期记忆中
     data = {
         "model": model,
         "messages": [
@@ -66,6 +73,9 @@ def get_openai_response(
     response.raise_for_status()
     response_json = response.json()
     st.session_state[session_keys["text_response"]] = response_json["choices"][0]["message"]["content"].strip()
+    st.session_state[session_keys["short_term_memory"]].append(
+        {"role": "assistant", "content": st.session_state[session_keys["text_response"]]}
+    )  # 添加助手响应到短期记忆中
     return response_json["choices"][0]["message"]["content"].strip()
 
 

src/chatbot/pages/setting.py

@@ -163,16 +163,16 @@ def message_box(title: str, message: str):
                 settings.cache_dir = cache_dir
                 settings.system_platform = system_platform  # type: ignore
                 settings.promopt = promopt
-                write_settings_file(settings_name="global.toml", settings=settings)
-                message_box("保存成功！", "你也可以通过手动配置 `global.toml` 来修改配置。")
+                write_settings_file(settings_name="config.toml", settings=settings)
+                message_box("保存成功！", "你也可以通过手动配置 `config.toml` 来修改配置。")
                 st.session_state[session_keys["initial_settings"]] = (
                     current_settings  # Update initial settings after save
                 )
             else:
                 message_box("未检测到更改", "配置未发生任何变化，无需保存。")
 
         if st.button("**恢复默认设置**", type="secondary", use_container_width=True):
-            settings = Path("config") / "global.toml"
+            settings = Path("config") / "config.toml"
             settings.unlink()
             load_settings_file("config.toml", ServiceSettings)
             message_box("恢复成功！", "配置已恢复为默认设置。刷新页面即可查看更改。")