File size: 4,495 Bytes
e83f85a 76a6e88 b28bc26 76a6e88 b28bc26 76a6e88 b28bc26 92c5d55 76a6e88 b28bc26 76a6e88 b28bc26 76a6e88 e8bc467 b28bc26 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 |
import torch
import pandas as pd
import numpy as np
import gradio as gr
import re
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
import re
from huggingface_hub import login
import os
from threading import Thread
# HF_TOKEN
TOKEN = os.getenv('HF_AUTH_TOKEN')
login(token=TOKEN,
add_to_git_credential=False)
# Open ai api key
API_KEY = os.getenv('OPEN_AI_API_KEY')
DESCRIPTION = '''
<div>
<h1 style="text-align: center;">Amphisbeana 🐍</h1>
<p>This uses Llama 3 and GPT-4o as generation, both of these make the final generation. <a href="https://huggingface.co/meta-llama/Meta-Llama-3-8B"><b>Llama3-8b</b></a>and <a href="https://platform.openai.com/docs/models/gpt-4o"><b>GPT-4o</b></a></p>
</div>
'''
# Place transformers in hardware to prepare for process and generation
llama_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B")
llama_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B", token=TOKEN, torch_dtype=torch.float16).to('cuda')
terminators = [
llama_tokenizer.eos_token_id,
llama_tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
# Place just input pass and return generation output
def llama_generation(input_text: str,
history: list,
temperature: float,
max_new_tokens: int):
"""
Pass input texts, tokenize, output and back to text.
"""
conversation = []
for user, assistant in history:
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
conversation.append({"role": "user", "content": input_text})
input_ids = llama_tokenizer.apply_chat_template(conversation, return_tensors='pt').to(llama_model.device)
# Skip_prompt, ignores the prompt in the chatbot
streamer = TextIteratorStreamer(llama_tokenizer, skip_prompt=True, skip_special_tokens=True)
# generation arguments to pass in llm generate() eventually
generate_kwargs = dict(
input_ids=input_ids,
streamer=streamer,
max_new_tokens=max_new_tokens,
do_sample=True,
temperature=temperature,
eos_token_id=terminators
)
# This makes a greedy generation when temperature is passed to 0 (selects the next token sequence generated by model regardless). Selects each token with the highest probability
if temperature == 0:
generate_kwargs["do_sample"] = False
# In order to use the generate_kwargs we need to place it in a thread which can also allow the UI to run different commands even when the model is generating
# place the function as target and place the kwargs next as the kwargs
thread = Thread(target=llama_model.generate, kwargs=generate_kwargs)
thread.start()
outputs = []
for text in streamer:
outputs.append(text)
return "".join(outputs)
# Let's just make sure the llama is returning as it should and than place that return output into a function making it fit into a base
# Prompt for gpt-4o
chatbot=gr.Chatbot(height=600, label="Amphisbeana AI")
with gr.Blocks(fill_height=True) as demo:
gr.Markdown(DESCRIPTION)
gr.ChatInterface(
fn=llama_generation,
chatbot=chatbot,
fill_height=True,
# These will effect the parameters args and kwargs inside the llama_generation function, that the ui can interact with from the code
additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
additional_inputs=[
# Slider feature users can interactive to effect the temperature of model
gr.Slider(minimum=0,
maximum=1,
step=0.1,
value=0.95,
label="Temperature",
render=False),
# Sliding feature for the max tokens for generation on model
gr.Slider(minimum=128,
maximum=1500,
step=1,
value=512,
label="Max new tokens",
render=False),
],
examples=[
["Make a poem of batman inside willy wonka"],
["How can you a burrito with just flour?"],
["How was saturn formed in 3 sentences"],
["How does the frontal lobe effect playing soccer"],
],
cache_examples=False
)
if __name__ == "__main__":
demo.launch() |