Lyte's picture
Update app.py
b9c3e64 verified
raw
history blame contribute delete
No virus
2.69 kB
import os
import gradio as gr
import copy
from llama_cpp import Llama
from huggingface_hub import hf_hub_download
llm = Llama(
model_path=hf_hub_download(
repo_id=os.environ.get("REPO_ID", "Lyte/Llama-3.1-8B-Instruct-Reasoner-1o1_v0.3"),
filename=os.environ.get("MODEL_FILE", "unsloth.Q4_K_M.gguf"),
),
n_ctx=4096,
n_gpu_layers=-1,
)
# Updated training prompt
training_prompt = """<|start_header_id|>system<|end_header_id|>
You are a world-class AI system, capable of complex reasoning and reflection and correcting your mistakes. Reason through the query/question, and then provide your final response. If you detect that you made a mistake in your reasoning at any point, correct yourself.<|eot_id|><|start_header_id|>user<|end_header_id|>
{}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
{}"""
EOS_TOKEN = "<|eot_id|>"
def generate_text(
message,
history: list[tuple[str, str]],
max_tokens,
temperature,
top_p,
):
temp = ""
input_prompt = ""
for user_input, assistant_response in history:
input_prompt += training_prompt.format(user_input, assistant_response)
input_prompt += training_prompt.format(message, "")
output = llm(
input_prompt,
temperature=temperature,
top_p=top_p,
top_k=40,
repeat_penalty=1.1,
max_tokens=max_tokens,
stop=[
EOS_TOKEN,
"<|endoftext|>"
],
stream=True,
)
# Stream and yield the response
for out in output:
stream = copy.deepcopy(out)
temp += stream["choices"][0]["text"]
yield temp
demo = gr.ChatInterface(
generate_text,
title="Llama-3.1-8B-Instruct-Reasoner",
description="Running LLM with https://github.com/abetlen/llama-cpp-python",
examples=[
['How to setup a human base on Mars? Give short answer.'],
['Explain theory of relativity to me like I’m 8 years old.'],
['What is 9,000 * 9,000?'],
['Write a pun-filled happy birthday message to my friend Alex.'],
['Justify why a penguin might make a good king of the jungle.']
],
cache_examples=False,
retry_btn=None,
undo_btn="Delete Previous",
clear_btn="Clear",
additional_inputs=[
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.95,
step=0.05,
label="Top-p (nucleus sampling)",
),
],
)
if __name__ == "__main__":
demo.launch()