Spaces:
Sleeping
Sleeping
File size: 5,299 Bytes
4043b0f 8191a3f 4043b0f a5ac299 4043b0f a5ac299 4043b0f a5ac299 4043b0f f54e5c5 4043b0f a5ac299 4043b0f f54e5c5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 |
# @title Think Paraguayo
import os
import random
import time
os.system("pip install gradio, llama_index, ragatouille, llama-cpp-python")
os.system("git clone https://github.com/EnPaiva93/think-paraguayo-space-aux.git")
os.system("wget https://huggingface.co/thinkPy/gua-a_v0.2-dpo_mistral-7b_GGUF/resolve/main/gua-a_v0.2-dpo_mistral-7b_q4_K_M.gguf -O model.gguf")
from llama_cpp import Llama
import gradio as gr
from ragatouille import RAGPretrainedModel
from llama_index.core import Document, SimpleDirectoryReader
from llama_index.core.node_parser import SentenceSplitter
max_seq_length = 512 # Choose any! We auto support RoPE Scaling internally!
prompt = """Responde a preguntas de forma clara, amable, concisa y solamente en el lenguaje español, sobre el libro Ñande Ypykuéra.
Contexto
-------------------------
{}
-------------------------
### Pregunta:
{}
### Respuesta:
{}"""
# Initialize the LLM
llm = Llama(model_path="model.gguf",
n_ctx=512,
n_threads=2)
BASE_PATH = "/home/user/app/think-paraguayo-space-aux/"
DOC_PATH = BASE_PATH+"index"
print(os.listdir())
documents = SimpleDirectoryReader(input_files=[BASE_PATH+"libro.txt"]).load_data()
parser = SentenceSplitter(chunk_size=128, chunk_overlap=64)
nodes = parser.get_nodes_from_documents(
documents, show_progress=False
)
list_nodes = [node.text for node in nodes]
print(os.getcwd())
if os.path.exists(DOC_PATH):
RAG = RAGPretrainedModel.from_index(DOC_PATH)
else:
RAG = RAGPretrainedModel.from_pretrained("AdrienB134/ColBERTv2.0-spanish-mmarcoES")
my_documents = list_nodes
index_path = RAG.index(index_name=DOC_PATH, max_document_length= 100, collection=my_documents)
# def convert_list_to_dict(lst):
# res_dct = {i: lst[i] for i in range(len(lst))}
# return res_dct
def reformat_rag(results_rag):
if results_rag is not None:
return [result["content"] for result in results_rag]
else:
return [""]
# def response(query: str = "Quien es gua'a?", context: str = ""):
# # print(base_prompt.format(query,""))
# inputs = tokenizer([base_prompt.format(query,"")], return_tensors = "pt").to("cuda")
# outputs = model.generate(**inputs, max_new_tokens = 128, temperature = 0.1, repetition_penalty=1.15, pad_token_id=tokenizer.eos_token_id)
# return tokenizer.batch_decode(outputs[0][inputs["input_ids"].shape[1]:].unsqueeze(0), skip_special_tokens=True)[0]
def chat_stream_completion(message, history):
context = reformat_rag(RAG.search(message, k=1))
context = " \n ".join(context)
full_prompt = prompt.format(context,message,"")
print(full_prompt)
response = llm.create_completion(
prompt=full_prompt,
temperature=0.01,
max_tokens=256,
stream=True
)
# print(response)
message_repl = ""
for chunk in response:
if len(chunk['choices'][0]["text"]) != 0:
# print(chunk)
message_repl = message_repl + chunk['choices'][0]["text"]
yield message_repl
# def answer_question(pipeline, character, question):
# def answer_question(question):
# # context = reformat_rag(RAG.search(question, k=2))
# # context = " \n ".join(context)
# yield chat_stream_completion(question, None)
# def answer_question(question):
# context = reformat_rag(RAG.search(question, k=2))
# context = " \n ".join(context)
# return response(question, "")
# def random_element():
# return random.choice(list_nodes)
# clear_output()
print("Importación Completada.. OK")
css = """
h1 {
font-size: 32px;
text-align: center;
}
h2 {
text-align: center;
}
img {
height: 750px; /* Reducing the image height */
}
"""
def launcher():
with gr.Blocks(css=css) as demo:
gr.Markdown("# Think Paraguayo")
gr.Markdown("## Conoce la cultura guaraní!!")
with gr.Row(variant='panel'):
with gr.Column(scale=1):
gr.Image(value=BASE_PATH+"think_paraguayo.jpeg", type="filepath", label="Imagen Estática")
with gr.Column(scale=1):
# with gr.Row():
# button = gr.Button("Cuentame ...")
# with gr.Row():
# textbox = gr.Textbox(label="", interactive=False, value=random_element())
# button.click(fn=random_element, inputs=[], outputs=textbox)
# with gr.Row():
chatbot = gr.ChatInterface(
fn=chat_stream_completion,
retry_btn = None,
stop_btn = None,
undo_btn = None
).queue()
# with gr.Row():
# msg = gr.Textbox()
# with gr.Row():
# clear = gr.ClearButton([msg, chatbot])
# def respond(message, chat_history):
# bot_message = answer_question(message)
# print(bot_message)
# chat_history.append((message, bot_message))
# time.sleep(2)
# return "", chat_history
# msg.submit(chat_stream_completion, [msg, chatbot], [msg, chatbot])
demo.launch(share=True, inline= False, debug=True)
if __name__ == "__main__":
launcher() |