ADD: initial implementation of ChatSeek using DeepSeek with Gradio interface

README.md

@@ -1,12 +1,12 @@
 ---
 title: ChatSeek
-emoji: 🐨
+emoji: 🐳
 colorFrom: blue
 colorTo: gray
 sdk: gradio
 sdk_version: 5.13.2
-app_file: app.py
-pinned: false
+app_file: main.py
+pinned: true
 license: apache-2.0
 short_description: Chat interface to use DeepSeekV3 and DeepSeekR1
 ---

main.py

@@ -0,0 +1,170 @@
+import gradio
+from huggingface_hub import InferenceClient
+
+import datetime
+import uuid
+import json
+
+import re
+
+import os
+import sys
+
+history = [
+    {"role": "system", "content": ""},
+]
+
+tmp_dir = os.getenv('TMP') or os.getenv('TMPDIR')
+
+def generate_uuid():
+    _uuid = datetime.datetime.now().strftime("%Y%m%d%H%M%S") + str(uuid.uuid4())
+    return _uuid
+
+def write_current_chat_to_file(current_chat):
+    with open(f"{tmp_dir}/{current_chat['chat_id']}.json", "w") as f:
+        json.dump(current_chat, f, indent = 4)
+
+def process_input_message(message_box, current_chat):
+    current_chat["chat_history"].append({"role": "user", "content": message_box["text"]})
+    return current_chat
+
+def get_text_between_tags(text, start_tag, end_tag):
+    pattern = rf'{re.escape(start_tag)}(.*?){re.escape(end_tag)}'
+    match = re.search(pattern, text, re.DOTALL)
+    return match.group(1) if match else ""
+
+def remove_text_between_tags(text, start_tag, end_tag):
+    pattern = rf'{re.escape(start_tag)}.*?{re.escape(end_tag)}'
+    return re.sub(pattern, '', text, flags=re.DOTALL)
+
+def call_chatbot(api_token, current_chat, system_message, max_tokens, temperature, top_p):
+    client = InferenceClient(
+        provider = "hf-inference",
+        api_key = api_token
+    )
+
+    current_chat["chat_history"][0]["content"] = system_message
+    current_chat["chat_history"].append({"content": "", "role": "assistant", "metadata": {}})
+    
+    stream = client.chat.completions.create(
+        model = "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", 
+        messages = current_chat["chat_history"], 
+        max_tokens = max_tokens,
+        temperature = temperature,
+        top_p = top_p,
+        stream = True
+    )
+
+    response = ""
+    for chunk in stream:
+        response = response + chunk.choices[0].delta.content
+        current_chat["chat_history"][-1]["content"] = response
+        yield current_chat["chat_history"], current_chat
+        #yield current_chat["chat_history"], current_chat
+    
+    current_chat["chat_history"][-1]["metadata"]["title"] = "💭 Thoughts:"
+    current_chat["chat_history"][-1]["content"] = get_text_between_tags(response, "<think>", "</think>")
+    current_chat["chat_history"].append({"content": remove_text_between_tags(response, "<think>", "</think>"), "role": "assistant"})
+    
+    yield current_chat["chat_history"], current_chat
+    
+def reset_chat():
+    chatbot = gradio.Chatbot(history, type = "messages")
+    message_box = gradio.MultimodalTextbox(
+        value = "",
+        interactive = True,
+        file_count = "multiple",
+        placeholder = "Enter message...",
+        show_label = False,
+        sources = [],
+        stop_btn = True,
+    )
+    
+    current_chat_id = generate_uuid()
+    current_chat = gradio.JSON(
+        {
+            "version": "0.1",
+            "chat_id": current_chat_id, 
+            "chat_history": history
+        }
+    )
+    save_chat_button = gradio.DownloadButton(label = "Save", value = f"{tmp_dir}/{current_chat_id}.json")
+    
+    system_message = gradio.Textbox(label = "System Message", value = "You are a helpful bot. Be concise with your answers. Do not think with more than 3 lines. Answer in 2 lines. Only answer in English")
+    max_tokens = gradio.Slider(label = "Max Tokens", minimum = 500, maximum = 3000, step = 100, value = 1000)
+    temperature = gradio.Slider(label = "Temperature", minimum = 0.1, maximum = 2.0, step = 0.1, value = 0.5)
+    top_p = gradio.Slider(label = "Top P", minimum = 0.1, maximum = 1.0, step = 0.1, value = 0.9)
+    
+    return chatbot, message_box, current_chat, save_chat_button, system_message, max_tokens, temperature, top_p
+
+def process_token(secret_token):
+    try:
+        passwords = os.environ.get("PASSWORDS")
+        passwords = passwords.split(":")
+        
+        if secret_token in passwords:
+            secret_token = os.environ.get("HF_KEY")
+        
+        return secret_token
+    except:
+        return secret_token
+    
+    
+    
+with gradio.Blocks(fill_height = True) as base_app:
+    gradio.Markdown("# ChatSeek")
+    gradio.Markdown("## ")
+    
+    with gradio.Row():
+        with gradio.Column(scale = 2):
+            secret_token = gradio.Textbox(label = "API Key", placeholder = "Enter Password/API Token. The key is never stored.", type = "password")
+        with gradio.Column():
+            submit_token = gradio.Button(value = "Authenticate", scale = 10)
+    
+    submit_token_call = submit_token.click(process_token, [secret_token], [secret_token])
+    
+    chatbot = gradio.Chatbot(history, type = "messages")
+    message_box = gradio.MultimodalTextbox(
+        interactive = True,
+        file_count = "multiple",
+        placeholder = "Enter message...",
+        show_label = False,
+        sources = [],
+        stop_btn = True,
+    )
+    
+    current_chat_id = generate_uuid()
+    
+    with gradio.Row():
+        with gradio.Column():
+            reset_button = gradio.Button(value = "Start a New Chat")
+        with gradio.Column():
+            save_chat_button = gradio.DownloadButton(label = "Save", value = f"{tmp_dir}/{current_chat_id}.json")
+            
+    with gradio.Accordion(label = "Advanced Parameters", open = False):
+        system_message = gradio.Textbox(label = "System Message", value = "You are a helpful bot. Be concise with your answers. Do not think with more than 3 lines. Answer in 2 lines. Only answer in English.")
+        max_tokens = gradio.Slider(label = "Max Tokens", minimum = 500, maximum = 3000, step = 100, value = 1000)
+        temperature = gradio.Slider(label = "Temperature", minimum = 0.1, maximum = 2.0, step = 0.1, value = 0.5)
+        top_p = gradio.Slider(label = "Top P", minimum = 0.1, maximum = 1.0, step = 0.1, value = 0.9)
+    
+    with gradio.Accordion(label = "Metadata", open = False):
+        current_chat = gradio.JSON(
+            {
+                "version": "0.1",
+                "chat_id": current_chat_id, 
+                "chat_history": history
+            },
+            visible = True
+        )
+    
+    submit_message_call = message_box.submit(process_input_message, [message_box, current_chat], [current_chat], queue=False).then(write_current_chat_to_file, [current_chat], [])
+    clear_message_box_call = submit_message_call.then(lambda: gradio.MultimodalTextbox(value = "", interactive = True) , None, [message_box])
+    invoke_chatbot_call = clear_message_box_call.then(call_chatbot, [secret_token, current_chat, system_message, max_tokens, temperature, top_p], [chatbot, current_chat]).then(write_current_chat_to_file, [current_chat], [])
+    
+    reset_button_call = reset_button.click(reset_chat, [], [chatbot, message_box, current_chat, save_chat_button, system_message, max_tokens, temperature, top_p])
+    
+if __name__ == "__main__":
+    base_app.launch(
+        allowed_paths = [tmp_dir]
+    )
+