Update app.py

app.py

@@ -28,7 +28,11 @@ from huggingface_hub import snapshot_download
 # @@ environments ================
 
 DEBUG = bool(int(os.environ.get("DEBUG", "1")))
-BLOCK_ZH = bool(int(os.environ.get("BLOCK_ZH", "1")))
+
+# List of languages to block
+BLOCK_LANGS = str(os.environ.get("BLOCK_LANGS", ""))
+BLOCK_LANGS = BLOCK_LANGS.strip().split(";") if len(BLOCK_LANGS.strip()) > 0 else []
+
 # for lang block, wether to block in history too
 LANG_BLOCK_HISTORY = bool(int(os.environ.get("LANG_BLOCK_HISTORY", "0")))
 TENSOR_PARALLEL = int(os.environ.get("TENSOR_PARALLEL", "1"))
@@ -64,6 +68,9 @@ STREAM_YIELD_MULTIPLE = int(os.environ.get("STREAM_YIELD_MULTIPLE", "1"))
 # how many iterations to perform safety check on response
 STREAM_CHECK_MULTIPLE = int(os.environ.get("STREAM_CHECK_MULTIPLE", "0"))
 
+# whether to enable to popup accept user
+ENABLE_AGREE_POPUP = bool(int(os.environ.get("ENABLE_AGREE_POPUP", "0")))
+
 # self explanatory
 MAX_TOKENS = int(os.environ.get("MAX_TOKENS", "2048"))
 TEMPERATURE = float(os.environ.get("TEMPERATURE", "0.1"))
@@ -88,7 +95,6 @@ MODEL_PATH=./seal-13b-chat-a
 
 """
 
-
 # ==============================
 print(f'DEBUG mode: {DEBUG}')
 print(f'Torch version: {torch.__version__}')
@@ -136,7 +142,6 @@ Your response should adapt to the norms and customs of the respective language a
 # ============ CONSTANT ============
 # https://github.com/gradio-app/gradio/issues/884
 MODEL_NAME = "SeaLLM-13B"
-MODEL_TITLE = "SeaLLM-13B - An Assistant for Southeast Asian Languages"
 
 MODEL_TITLE = """
 <div class="container" style="
@@ -212,22 +217,6 @@ path_markdown = """
 
 
 
-
-def _detect_lang(text):
-    # Disable language that may have safety risk
-    from langdetect import detect as detect_lang
-    dlang = None
-    try:
-        dlang = detect_lang(text)
-    except Exception as e:
-        print(f'Error: {e}')
-        if "No features in text." in str(e):
-            return "en"
-        else:
-            return "zh"
-    return dlang
-
-
 def custom_hf_model_weights_iterator(
     model_name_or_path: str,
     cache_dir: Optional[str] = None,
@@ -1003,18 +992,39 @@ Please also consider clearing the chat box for a better experience."""
 
 KEYWORD_BLOCK_MESSAGE = "Sorry, I cannot fulfill your request. If you have any unrelated question, I'll be glad to help."
 
-def block_zh(
+
+def _detect_lang(text):
+    # Disable language that may have safety risk
+    from langdetect import detect as detect_lang
+    dlang = None
+    try:
+        dlang = detect_lang(text)
+    except Exception as e:
+        print(f'Error: {e}')
+        if "No features in text." in str(e):
+            return "en"
+        else:
+            return "zh"
+    return dlang
+
+
+def block_lang(
     message: str, 
     history: List[Tuple[str, str]] = None,
 ) -> str:
     # relieve history base block
+    if len(BLOCK_LANGS) == 0:
+        return False
+    
     if LANG_BLOCK_HISTORY and history is not None and any((LANG_BLOCK_MESSAGE in x[1].strip()) for x in history):
         return True
-    elif 'zh' in _detect_lang(message):
-        print(f'Detect zh: {message}')
-        return True
     else:
-        return False
+        _lang = _detect_lang(message)
+        if _lang in BLOCK_LANGS:
+            print(f'Detect blocked {_lang}: {message}')
+            return True
+        else:
+            return False
 
 
 def log_responses(history, message, response):
@@ -1029,13 +1039,9 @@ def safety_check(text, history=None, ) -> Optional[str]:
     if len(KEYWORDS) > 0 and any(x in text.lower() for x in KEYWORDS):
         return KEYWORD_BLOCK_MESSAGE
     
-    if BLOCK_ZH:
-        if history is not None:
-            if block_zh(text, history):
-                return LANG_BLOCK_MESSAGE
-        else:
-            if "zh" in _detect_lang(text):
-                return LANG_BLOCK_MESSAGE
+    if len(BLOCK_LANGS) > 0:
+        if block_lang(text, history):
+            return LANG_BLOCK_MESSAGE
 
     return None
 
@@ -1159,6 +1165,7 @@ def check_model_path(model_path) -> str:
 
 def maybe_delete_folder():
     if IS_DELETE_FOLDER and DOWNLOAD_SNAPSHOT:
+        import shutil
         print(f'DELETE ALL FILES IN {DELETE_FOLDER}')
         for filename in os.listdir(DELETE_FOLDER):
             file_path = os.path.join(DELETE_FOLDER, filename)
@@ -1170,6 +1177,11 @@ def maybe_delete_folder():
             except Exception as e:
                 print('Failed to delete %s. Reason: %s' % (file_path, e))
 
+AGREE_POP_SCRIPTS = """
+async () => {
+    alert("To use our service, you are required to agree to the following terms:\\nYou must not use our service to generate any harmful, unethical or illegal content that violates local and international laws, including but not limited to hate speech, violence and deception.\\nThe service may collect user dialogue data for performance improvement, and reserves the right to distribute it under CC-BY or similar license. So do not enter any personal information!");
+}
+"""
 
 def launch():
     global demo, llm, DEBUG
@@ -1187,8 +1199,10 @@ def launch():
     ckpt_info = "None"
 
     print(
-        f'Launch config: {tensor_parallel=} / {dtype=} / {max_tokens} | {BLOCK_ZH=} '
+        f'Launch config: {tensor_parallel=} / {dtype=} / {max_tokens} '
         f'\n| model_title=`{model_title}` '
+        f'\n| BLOCK_LANGS={BLOCK_LANGS} '
+        f'\n| IS_DELETE_FOLDER={IS_DELETE_FOLDER} '
         f'\n| STREAM_YIELD_MULTIPLE={STREAM_YIELD_MULTIPLE} '
         f'\n| STREAM_CHECK_MULTIPLE={STREAM_CHECK_MULTIPLE} '
         f'\n| DISPLAY_MODEL_PATH={DISPLAY_MODEL_PATH} '
@@ -1210,6 +1224,7 @@ def launch():
         print(f'Creating in DEBUG MODE')
     else:
         # ! load the model
+        maybe_delete_folder()
 
         if DOWNLOAD_SNAPSHOT:
             print(f'Downloading from HF_MODEL_NAME={hf_model_name} -> {model_path}')
@@ -1258,7 +1273,8 @@ def launch():
             latex_delimiters=[
                 { "left": "$", "right": "$", "display": False},
                 { "left": "$$", "right": "$$", "display": True},
-            ]
+            ],
+            show_copy_button=True,
         ),
         textbox=gr.Textbox(placeholder='Type message', lines=8, max_lines=128, min_width=200),
         submit_btn=gr.Button(value='Submit', variant="primary", scale=0),
@@ -1276,10 +1292,13 @@ def launch():
     )
     demo.title = MODEL_NAME
     with demo:
-        # gr.Markdown(warning_markdown)
         gr.Markdown(cite_markdown)
         if DISPLAY_MODEL_PATH:
             gr.Markdown(path_markdown.format(model_path=model_path))
+        
+        if ENABLE_AGREE_POPUP:
+            demo.load(None, None, None, _js=AGREE_POP_SCRIPTS)
+    
 
     demo.queue()
     demo.launch(server_port=PORT)