session satte warp

docker-compose.yml

@@ -8,13 +8,15 @@ services:
       options:
         max-size: "50m"
     environment:
-        - API_KEY=${API_KEY}
+        - GPTS_API_SERVER=${GPTS_API_SERVER}
+        - API_SECRET=${API_KEY}
         - OPENAI_API_TYPE=${OPENAI_API_TYPE}
         - AZURE_OPENAI_API_VERSION=${AZURE_OPENAI_API_VERSION}
         - AZURE_OPENAI_API_BASE=${AZURE_OPENAI_API_BASE}
         - AZURE_OPENAI_API_KEY=${AZURE_OPENAI_API_KEY}
         - OPENAI_API_KEY=${OPENAI_API_KEY}
         - QDRANT_URL=${QDRANT_URL}
+        - QDRANT_KEY=${QDRANT_KEY}
         - DATA_DIR=/data
     volumes:
       - gptservice-volume:/data
@@ -32,12 +34,15 @@ services:
       options:
         max-size: "50m"
     environment:
+        - GPT_SERVICE_ADDRESS=${GPT_SERVICE_ADDRESS}
+        - GPT_SERVICE_TOKEN=${GPT_SERVICE_TOKEN}
         - OPENAI_API_TYPE=${OPENAI_API_TYPE}
+        - OPENAI_API_KEY=${OPENAI_API_KEY}
         - AZURE_OPENAI_API_VERSION=${AZURE_OPENAI_API_VERSION}
         - AZURE_OPENAI_API_BASE=${AZURE_OPENAI_API_BASE}
         - AZURE_OPENAI_API_KEY=${AZURE_OPENAI_API_KEY}
-        - OPENAI_API_KEY=${OPENAI_API_KEY}
-        - QDRANT_URL=${QDRANT_URL}
+        - MSAL_TENANTID=${MSAL_TENANTID}
+        - MSAL_APPID=${MSAL_APPID}
         - DATA_DIR=/data
     volumes:
       - gptstudio-volume:/data

libs/session.py

@@ -0,0 +1,29 @@
+import streamlit as st
+
+
+class PageSessionState:
+    def __init__(self, prefix):
+        self._prefix = prefix
+
+    def initn_attr(self, key: str, default_value: object):
+        if not hasattr(self, key):
+            setattr(self, key, default_value)
+
+    def __getattr__(self, key):
+        if key == "_prefix":
+            return self.__dict__[key]
+        return st.session_state.get(f"{self._prefix}_{key}", None)
+
+    def __setattr__(self, key, value):
+        if key == "_prefix":
+            self.__dict__[key] = value
+        else:
+            st.session_state[f"{self._prefix}_{key}"] = value
+
+    def __delattr__(self, key):
+        if key == "_prefix":
+            raise AttributeError("Cannot delete _prefix attribute")
+        st.session_state.pop(f"{self._prefix}_{key}", None)
+
+    def newkey(self, key):
+        return f"{self._prefix}_{key}"

pages/04_Speech_Transcribe.py

@@ -11,38 +11,51 @@
 import os
 from dotenv import load_dotenv
 
+from libs.session import PageSessionState
+
 sys.path.append(os.path.abspath('..'))
 load_dotenv()
 
+page_state = PageSessionState("speech_transcribe")
+
 with st.sidebar:
     value = msal_auth()
     if value is None:
         st.stop()
 
+st.sidebar.markdown("# 🎙️语音转录🎤")
 
 st.markdown("# 🎙️语音转录🎤")
 st.markdown("> 上传文本或者录制语音识别，然后合成新的语音")
 
-if "audio_recode" not in st.session_state:
-    st.session_state.audio_recode = None
-
-if "speech_recode" not in st.session_state:
-    st.session_state.speech_recode = None
-
-if "audio_processing" not in st.session_state:
-    st.session_state.audio_processing = False
+# 音频录制内容
+page_state.initn_attr("audio_recode", None)
+# 语音合成内容
+page_state.initn_attr("speech_recode", None)
+# 是否正在处理中
+page_state.initn_attr("audio_processing", False)
 
+# 用于存储临时文件
 data_dir = os.getenv("DATA_DIR", "/tmp/data")
 if not os.path.exists(data_dir):
     os.makedirs(data_dir)
 
-uploaded_file = st.file_uploader("上传文本文件", type=["txt", "md"])
+content_box = st.empty()
+
+uploaded_file = st.sidebar.file_uploader("上传文本文件", type=["txt", "md"])
 if uploaded_file is not None:
     stringio = io.StringIO(uploaded_file.getvalue().decode("utf-8"))
     string_data = stringio.read()
-    st.session_state.audio_recode = string_data
+    page_state.audio_recode = string_data
+
+if st.sidebar.button("录制音频"):
+    page_state.audio_recode = None
+    page_state.audio_processing = False
+    page_state.speech_recode = None
+    content_box.empty()
+    st.rerun()
 
-if st.session_state.audio_recode is None:
+if page_state.audio_recode is None:
     with st.spinner('正在识别语音...'):
         wav_audio_recode = audio_recorder("点击录音", icon_size="2x", pause_threshold=3.0)
         if wav_audio_recode is not None:
@@ -56,12 +69,11 @@
                 response_format="json",
                 file=open(filename, "rb"),
             )
-            st.session_state.audio_recode = transcript.text
+            page_state.audio_recode = transcript.text
             st.rerun()
 
-if st.session_state.audio_recode is not None:
-    st.markdown("### 🎤语音合成")
-    st.markdown(st.session_state.audio_recode)
+if page_state.audio_recode is not None:
+    content_box.markdown(page_state.audio_recode)
     sound = st.selectbox("选择音色", ["alloy", "echo", "fable", "onyx", "nova", "shimmer"])
     c1, c2, c3 = st.columns(3)
     if c1.button("合成语音"):
@@ -71,26 +83,20 @@
             response = client.audio.speech.create(
                 model="tts-1",
                 voice=sound,
-                input=st.session_state.audio_recode
+                input=page_state.audio_recode
             )
-            st.session_state.speech_recode = response.read()
+            page_state.speech_recode = response.read()
             st.write(f"🎧{sound}音色")
-            st.audio(st.session_state.speech_recode, format="audio/mp3")
+            st.audio(page_state.speech_recode, format="audio/mp3")
             st.write(f"语音{sound}合成完成")
             status.update(label="语音合成完成!", state="complete")
 
-    if c2.button("重新录制"):
-        st.session_state.audio_recode = None
-        st.session_state.audio_processing = False
-        st.session_state.speech_recode = None
-        st.rerun()
-
-    if st.session_state.speech_recode is not None:
+    if page_state.speech_recode is not None:
         c3.download_button(
             label="下载语音",
-            data=st.session_state.speech_recode,
+            data=page_state.speech_recode,
             file_name='speech.mp3',
         )
 else:
-    if st.session_state.audio_processing:
+    if page_state.audio_processing:
         st.write("还有任务在处理")

pages/05_Video.py

@@ -1,5 +1,25 @@
-from streamlit_webrtc import webrtc_streamer
+import cv2
+import numpy as np
+import streamlit as st
+from camera_input_live import camera_input_live
 
+"# Streamlit camera input live Demo"
+"## Try holding a qr code in front of your webcam"
 
+image = camera_input_live(debounce=2000, width=360, height=240, show_controls=True)
 
-webrtc_streamer(key="sample")
+if image is not None:
+    st.image(image)
+    bytes_data = image.getvalue()
+    cv2_img = cv2.imdecode(np.frombuffer(bytes_data, np.uint8), cv2.IMREAD_COLOR)
+
+    detector = cv2.QRCodeDetector()
+
+    data, bbox, straight_qrcode = detector.detectAndDecode(cv2_img)
+
+    if data:
+        st.write("# Found QR code")
+        st.write(data)
+        with st.expander("Show details"):
+            st.write("BBox:", bbox)
+            st.write("Straight QR code:", straight_qrcode)

pages/06_Image_Analysis.py

@@ -0,0 +1,109 @@
+import streamlit as st
+from libs.llms import openai_analyze_image, openai_streaming
+from libs.msal import msal_auth
+from libs.session import PageSessionState
+
+st.set_page_config(page_title="视觉分析", page_icon="🔬")
+
+page_state = PageSessionState("image_analysis")
+
+with st.sidebar:
+    value = msal_auth()
+    if value is None:
+        st.stop()
+
+
+# 用于存储对话记录
+page_state.initn_attr("messages", [])
+
+# 用于标记上一条用户消息是否已经处理
+page_state.initn_attr("last_user_msg_processed", True)
+
+# 用于存储图像分析结果
+page_state.initn_attr("analysis_result", "")
+
+page_state.initn_attr("input_type", "camera")
+
+st.sidebar.markdown("# 🔬视觉分析")
+
+st.title("🔬视觉分析")
+
+# 图像分析提示输入
+prompt = st.sidebar.text_area("图像分析提示", "识别分析图片内容", height=40)
+
+
+def clear_result():
+    page_state.analysis_result = ""
+    page_state.last_user_msg_processed = True
+    page_state.messages = []
+    if page_state.input_type == "camera" and page_state.camera_image is not None:
+        with st.spinner("分析中..."):
+            page_state.analysis_result = openai_analyze_image(prompt, page_state.camera_image)
+            page_state.messages.append({"role": "assistant", "content": page_state.analysis_result})
+
+
+# 摄像头输入获取图片
+if st.sidebar.selectbox("选择图片输入方式", ["摄像头", "上传图片"]) == "摄像头":
+    page_state.input_type = "camera"
+    image = st.camera_input("点击按钮截图", on_change=clear_result, key="image_analysis_camera_image")
+else:
+    page_state.input_type = "upload"
+    image = st.sidebar.file_uploader("上传图片", type=["png", "jpg", "jpeg"],
+                             on_change=clear_result, key="image_analysis_camera_image")
+
+if page_state.camera_image is not None:
+    if page_state.input_type == "upload":
+        st.image(page_state.camera_image, caption="上传的图片", use_column_width=True)
+        c1, c2 = st.columns(2)
+        if c1.button("分析图像"):
+            with st.spinner("分析中..."):
+                page_state.analysis_result = openai_analyze_image(prompt, page_state.camera_image)
+                page_state.messages.append({"role": "assistant", "content": page_state.analysis_result})
+
+        if c2.button("清除结果"):
+            clear_result()
+
+# 设置对话记录
+for msg in page_state.messages:
+    with st.chat_message(msg["role"]):
+        st.write(msg["content"])
+
+# 输入用户消息
+if uprompt := st.chat_input("输入你的问题"):
+    # 用于标记用户消息还没有处理
+    page_state.last_user_msg_processed = False
+    page_state.messages.append({"role": "user", "content": uprompt})
+    with st.chat_message("user"):
+        st.write(uprompt)
+
+# 用户输入响应，如果上一条消息不是助手的消息，且上一条用户消息还没有处理完毕
+if ((page_state.messages
+     and page_state.messages[-1]["role"] != "assistant"
+     and not page_state.last_user_msg_processed)
+        and page_state.analysis_result not in [""]):
+    # 处理响应
+    with st.chat_message("assistant"):
+        with st.spinner("Thinking..."):
+            sysmsg = f""""
+            以下是来自一图片识别获取的内容结果：
+            '''
+            {page_state.analysis_result}
+            '''
+            我们将围绕这个内容进行深入讨论。
+            """
+            response = openai_streaming(sysmsg, page_state.messages[-10:])
+            # 流式输出
+            placeholder = st.empty()
+            full_response = ''
+            for item in response:
+                text = item.content
+                if text is not None:
+                    full_response += text
+                    placeholder.markdown(full_response)
+            placeholder.markdown(full_response)
+
+    # 用于标记上一条用户消息已经处理完毕
+    page_state.last_user_msg_processed = True
+    # 追加对话记录
+    message = {"role": "assistant", "content": full_response}
+    page_state.messages.append(message)