Update app.py

app.py

@@ -10,44 +10,31 @@ client = InferenceClient(model)
 # Embedded system prompt
 system_prompt_text = "You are a smart and helpful co-worker of Thailand based multi-national company PTT, and PTTEP. You help with any kind of request and provide a detailed answer to the question. But if you are asked about something unethical or dangerous, you must refuse and provide a safe and respectful way to handle that."
 
-# Read the content of the info.md and info2.md files
+# Read the content of the info.md file
 with open("info.md", "r") as file:
     info_md_content = file.read()
 
-with open("info2.md", "r") as file:
-    info2_md_content = file.read()
-
-# Chunk the info.md and info2.md content into smaller sections
-chunk_size = 1500  # Adjust this size as needed to fit the context window
+# Chunk the info.md content into smaller sections
+chunk_size = 2500  # Adjust this size as needed
 info_md_chunks = textwrap.wrap(info_md_content, chunk_size)
-info2_md_chunks = textwrap.wrap(info2_md_content, chunk_size)
-
-# Combine both sets of chunks
-all_chunks = info_md_chunks + info2_md_chunks
 
-# Function to initialize the conversation history with chunks
-def initialize_history(chunks):
-    history = []
-    for chunk in chunks:
-        history.append(("System Information", chunk))
-    return history
+def get_all_chunks(chunks):
+    return "\n\n".join(chunks)
 
-# Initialize history with initial chunks
-history = initialize_history(all_chunks[:2])  # Starting with the first two chunks for example
-
-def format_prompt_mixtral(message, history):
+def format_prompt_mixtral(message, history, info_md_chunks):
     prompt = "<s>"
+    all_chunks = get_all_chunks(info_md_chunks)
+    prompt += f"{all_chunks}\n\n"  # Add all chunks of info.md at the beginning
     prompt += f"{system_prompt_text}\n\n"  # Add the system prompt
 
-    # Include the initial context from the chunks
-    for user_prompt, bot_response in history:
-        prompt += f"[INST] {user_prompt} [/INST] {bot_response}</s> "
-
-    # Add the current user message
+    if history:
+        for user_prompt, bot_response in history:
+            prompt += f"[INST] {user_prompt} [/INST]"
+            prompt += f" {bot_response}</s> "
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
-def chat_inf(message, history, seed, temp, tokens, top_p, rep_p):
+def chat_inf(prompt, history, seed, temp, tokens, top_p, rep_p):
     generate_kwargs = dict(
         temperature=temp,
         max_new_tokens=tokens,
@@ -57,29 +44,27 @@ def chat_inf(message, history, seed, temp, tokens, top_p, rep_p):
         seed=seed,
     )
 
-    formatted_prompt = format_prompt_mixtral(message, history)
+    formatted_prompt = format_prompt_mixtral(prompt, history, info_md_chunks)
     stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
     output = ""
     for response in stream:
         output += response.token.text
-        yield [(message, output)]
-    history.append((message, output))
+        yield [(prompt, output)]
+    history.append((prompt, output))
     yield history
 
 def clear_fn():
-    global history
-    history = initialize_history(all_chunks[:2])  # Reset to initial chunks
     return None, None
 
 rand_val = random.randint(1, 1111111111111111)
 
-def check_rand(rand, val):
-    if rand:
-        return random.randint(1, 1111111111111111)
+def check_rand(inp, val):
+    if inp:
+        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=random.randint(1, 1111111111111111))
     else:
-        return int(val)
+        return gr.Slider(label="Seed", minimum=1, maximum=1111111111111111, value=int(val))
 
-with gr.Blocks() as app:
+with gr.Blocks() as app:  # Add auth here
     gr.HTML("""<center><h1 style='font-size:xx-large;'>PTT Chatbot</h1><br><h3>running on Huggingface Inference </h3><br><h7>EXPERIMENTAL</center>""")
     with gr.Row():
         chat = gr.Chatbot(height=500)
@@ -105,13 +90,7 @@ with gr.Blocks() as app:
 
     hid1 = gr.Number(value=1, visible=False)
 
-    def on_chat(message, chat, seed, temp, tokens, top_p, rep_p):
-        chat.clear()
-        history = initialize_history(all_chunks[:2])
-        for response in chat_inf(message, history, seed, temp, tokens, top_p, rep_p):
-            chat.append(*response)
-    
-    go = btn.click(on_chat, [inp, chat, seed, temp, tokens, top_p, rep_p], chat)
+    go = btn.click(check_rand, [rand, seed], seed).then(chat_inf, [inp, chat, seed, temp, tokens, top_p, rep_p], chat)
 
     stop_btn.click(None, None, None, cancels=[go])
     clear_btn.click(clear_fn, None, [inp, chat])