Update app.py
Browse files
app.py
CHANGED
@@ -4,7 +4,13 @@ from llama_cpp import Llama
|
|
4 |
llm = Llama(model_path="model.gguf", n_ctx=8000, n_threads=2, chat_format="chatml")
|
5 |
|
6 |
def generate(message, history,temperature=0.3,max_tokens=512):
|
7 |
-
system_prompt = """You are a super Inteligent AI assistant
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
formatted_prompt = [{"role": "system", "content": system_prompt}]
|
9 |
for user_prompt, bot_response in history:
|
10 |
formatted_prompt.append({"role": "user", "content": user_prompt})
|
@@ -23,7 +29,7 @@ avatar_images=["user.png", "botnb.png"], bubble_full_width=False, show_label=Fal
|
|
23 |
iface = gr.ChatInterface(fn=generate, chatbot=mychatbot, retry_btn=None, undo_btn=None)
|
24 |
|
25 |
with gr.Blocks() as demo:
|
26 |
-
gr.HTML("<center><h1>
|
27 |
iface.render()
|
28 |
|
29 |
demo.queue().launch(show_api=False, server_name="0.0.0.0")
|
|
|
4 |
llm = Llama(model_path="model.gguf", n_ctx=8000, n_threads=2, chat_format="chatml")
|
5 |
|
6 |
def generate(message, history,temperature=0.3,max_tokens=512):
|
7 |
+
system_prompt = """You are a super Inteligent AI assistant.
|
8 |
+
I want you to think smartly, step by step.
|
9 |
+
Once you've thought through things step by step, check the responses
|
10 |
+
before issuing them. I want you to answer clearly, accurately,
|
11 |
+
and without any unnecessary words. I want you to be concise and provide exact answers,
|
12 |
+
with known data, without making things up. You're called "Little Llama",
|
13 |
+
you're a language model that was compressed but you're still the smartest!"""
|
14 |
formatted_prompt = [{"role": "system", "content": system_prompt}]
|
15 |
for user_prompt, bot_response in history:
|
16 |
formatted_prompt.append({"role": "user", "content": user_prompt})
|
|
|
29 |
iface = gr.ChatInterface(fn=generate, chatbot=mychatbot, retry_btn=None, undo_btn=None)
|
30 |
|
31 |
with gr.Blocks() as demo:
|
32 |
+
gr.HTML("<center><h1>Llama 13b - GGUF Q_4_K_M</h1></center>")
|
33 |
iface.render()
|
34 |
|
35 |
demo.queue().launch(show_api=False, server_name="0.0.0.0")
|