File size: 15,747 Bytes
58023c8 6748e1c 58023c8 6748e1c 58023c8 6748e1c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 |
import torch
from transformers import (
LlamaForCausalLM,
LlamaTokenizer,
StoppingCriteria,
)
import gradio as gr
import argparse
import os
from queue import Queue
from threading import Thread
import traceback
import gc
import torch
from auto_gptq import AutoGPTQForCausalLM
from langchain import HuggingFacePipeline, PromptTemplate
from langchain.chains import RetrievalQA
from langchain.document_loaders import PyPDFDirectoryLoader, DirectoryLoader
from langchain.embeddings import HuggingFaceInstructEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma
from pdf2image import convert_from_path
from transformers import AutoTokenizer, TextStreamer, pipeline
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
# Parse command-line arguments
parser = argparse.ArgumentParser()
parser.add_argument(
'--base_model',
default=None,
type=str,
help='Base model path')
parser.add_argument('--lora_model', default=None, type=str,
help="If None, perform inference on the base model")
parser.add_argument(
'--tokenizer_path',
default=None,
type=str,
help='If None, lora model path or base model path will be used')
parser.add_argument(
'--gpus',
default="0",
type=str,
help='If None, cuda:0 will be used. Inference using multi-cards: --gpus=0,1,... ')
parser.add_argument('--share', default=True, help='Share gradio domain name')
parser.add_argument('--port', default=19324, type=int, help='Port of gradio demo')
parser.add_argument(
'--max_memory',
default=256,
type=int,
help='Maximum input prompt length, if exceeded model will receive prompt[-max_memory:]')
parser.add_argument(
'--load_in_8bit',
action='store_true',
help='Use 8 bit quantified model')
parser.add_argument(
'--only_cpu',
action='store_true',
help='Only use CPU for inference')
parser.add_argument(
'--alpha',
type=str,
default="1.0",
help="The scaling factor of NTK method, can be a float or 'auto'. ")
args = parser.parse_args()
if args.only_cpu is True:
args.gpus = ""
#from patches import apply_attention_patch, apply_ntk_scaling_patch
#apply_attention_patch(use_memory_efficient_attention=True)
#apply_ntk_scaling_patch(args.alpha)
# Set CUDA devices if available
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpus
# Peft library can only import after setting CUDA devices
from peft import PeftModel
# Set up the required components: model and tokenizer
def setup():
global tokenizer, model, device, share, port, max_memory, vector_store
max_memory = args.max_memory
port = args.port
share = args.share
load_in_8bit = args.load_in_8bit
load_type = torch.float16
if torch.cuda.is_available():
device = torch.device(0)
else:
device = torch.device('cpu')
"""
if args.tokenizer_path is None:
args.tokenizer_path = args.lora_model
if args.lora_model is None:
args.tokenizer_path = args.base_model
"""
#先读取embedding模型
embeddings = HuggingFaceInstructEmbeddings(
model_name="BAAI/bge-large-en-v1.5", model_kwargs={"device": DEVICE}
)
#如果之前没有本地的faiss仓库,就把doc读取到向量库后,再把向量库保存到本地
if os.path.exists("/home/ywang/db")==False:
#=======加载知识库=======
loader = DirectoryLoader("kb")
docs = loader.load()
# splitting pdf into chunks with size of 1024
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
texts = text_splitter.split_documents(docs)
vector_store = Chroma.from_documents(texts, embeddings, persist_directory="db")
#如果本地已经有faiss仓库了,说明之前已经保存过了,就直接读取
else:
vector_store=Chroma(persist_directory="db", embedding_function=embeddings)
model_name_or_path = "TheBloke/Llama-2-13B-chat-GPTQ"
model_basename = "model"
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)
base_model = AutoGPTQForCausalLM.from_quantized(
model_name_or_path,
revision="gptq-4bit-128g-actorder_True",
model_basename=model_basename,
use_safetensors=True,
trust_remote_code=True,
inject_fused_attention=False,
device=DEVICE,
quantize_config=None,
)
model_vocab_size = base_model.get_input_embeddings().weight.size(0)
tokenzier_vocab_size = len(tokenizer)
print(f"Vocab of the base model: {model_vocab_size}")
print(f"Vocab of the tokenizer: {tokenzier_vocab_size}")
if model_vocab_size != tokenzier_vocab_size:
assert tokenzier_vocab_size > model_vocab_size
print("Resize model embeddings to fit tokenizer")
base_model.resize_token_embeddings(tokenzier_vocab_size)
if args.lora_model is not None:
print("loading peft model")
model = PeftModel.from_pretrained(
base_model,
args.lora_model,
torch_dtype=load_type,
device_map='auto',
)
else:
model = base_model
if device == torch.device('cpu'):
model.float()
model.eval()
DEFAULT_SYSTEM_PROMPT = """
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
""".strip()
def generate_prompt(prompt: str, system_prompt: str = DEFAULT_SYSTEM_PROMPT) -> str:
return f"""
[INST] <<SYS>>
{system_prompt}
<</SYS>>
{prompt} [/INST]
""".strip()
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
text_pipeline = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=2048,
temperature=0,
top_p=0.95,
repetition_penalty=1.15,
streamer=streamer,
)
llm = HuggingFacePipeline(pipeline=text_pipeline, model_kwargs={"temperature": 0})
SYSTEM_PROMPT = "Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer."
template = generate_prompt(
"""
{context}
Question: {question}
""",
system_prompt=SYSTEM_PROMPT,
)
prompt = PromptTemplate(template=template, input_variables=["context", "question"])
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=vector_store.as_retriever(search_kwargs={"k": 2}),
return_source_documents=True,
chain_type_kwargs={"prompt": prompt},
)
# Reset the user input
def reset_user_input():
return gr.update(value='')
# Reset the state
def reset_state():
return []
# Generate the prompt for the input of LM model
"""
def generate_prompt(instruction,my_input):
return f"Instruction:{my_input}\n Response:{instruction}"
"""
# User interaction function for chat
def user(user_message, history):
return gr.update(value="", interactive=False), history + \
[[user_message, None]]
class Stream(StoppingCriteria):
def __init__(self, callback_func=None):
self.callback_func = callback_func
def __call__(self, input_ids, scores) -> bool:
if self.callback_func is not None:
self.callback_func(input_ids[0])
return False
class Iteratorize:
"""
Transforms a function that takes a callback
into a lazy iterator (generator).
Adapted from: https://stackoverflow.com/a/9969000
"""
def __init__(self, func, kwargs=None, callback=None):
self.mfunc = func
self.c_callback = callback
self.q = Queue()
self.sentinel = object()
self.kwargs = kwargs or {}
self.stop_now = False
def _callback(val):
if self.stop_now:
raise ValueError
self.q.put(val)
def gentask():
try:
ret = self.mfunc(callback=_callback, **self.kwargs)
except ValueError:
pass
except Exception:
traceback.print_exc()
clear_torch_cache()
self.q.put(self.sentinel)
if self.c_callback:
self.c_callback(ret)
self.thread = Thread(target=gentask)
self.thread.start()
def __iter__(self):
return self
def __next__(self):
obj = self.q.get(True, None)
if obj is self.sentinel:
raise StopIteration
else:
return obj
def __del__(self):
clear_torch_cache()
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.stop_now = True
clear_torch_cache()
def clear_torch_cache():
gc.collect()
if torch.cuda.device_count() > 0:
torch.cuda.empty_cache()
# Perform prediction based on the user input and history
@torch.no_grad()
def predict(
history,
max_new_tokens=128,
top_p=0.75,
temperature=0.1,
top_k=40,
do_sample=True,
repetition_penalty=1.0
):
history[-1][1] = ""
history[-1][1] = qa_chain(history[-1][0])['result']
"""
#history的格式:[[query1,response1],[query2,response2],[query3,response3]……]
docs=vector_store.similarity_search(history[-1][0])
context=[doc.page_content for doc in docs]
#使用下面的方式,把多轮对话转为单轮对话
input = f"### Instruction:{history[-1][0]} ### Response:{history[-1][1]}"
prompt = generate_prompt(input,"".join(context))
inputs = tokenizer(qa_chain, return_tensors="pt")
input_ids = inputs["input_ids"].to(device)
generate_params = {
'input_ids': input_ids,
'max_new_tokens': max_new_tokens,
'top_p': top_p,
'temperature': temperature,
'top_k': top_k,
'do_sample': do_sample,
'repetition_penalty': repetition_penalty,
}
def generate_with_callback(callback=None, **kwargs):
if 'stopping_criteria' in kwargs:
kwargs['stopping_criteria'].append(Stream(callback_func=callback))
else:
kwargs['stopping_criteria'] = [Stream(callback_func=callback)]
clear_torch_cache()
with torch.no_grad():
model.generate(**kwargs)
def generate_with_streaming(**kwargs):
return Iteratorize(generate_with_callback, kwargs, callback=None)
with generate_with_streaming(**generate_params) as generator:
for output in generator:
next_token_ids = output[len(input_ids[0]):]
if next_token_ids[0] == tokenizer.eos_token_id:
break
new_tokens = tokenizer.decode(
next_token_ids, skip_special_tokens=True)
if isinstance(tokenizer, LlamaTokenizer) and len(next_token_ids) > 0:
if tokenizer.convert_ids_to_tokens(int(next_token_ids[0])).startswith('▁'):
new_tokens = ' ' + new_tokens
history[-1][1] = new_tokens
yield history
if len(next_token_ids) >= max_new_tokens:
break
"""
yield history
# Call the setup function to initialize the components
setup()
# Create the Gradio interface
with gr.Blocks() as demo:
github_banner_path = 'https://radformation.com/images/radformation-logo-white.svg'
#gr.HTML(f'<p align="center"><a href="https://radformation.com/"><img src={github_banner_path} width="700"/></a></p>')
gr.Markdown("> Radformation Q&A bot")
chatbot = gr.Chatbot()
with gr.Row():
with gr.Column(scale=4):
with gr.Column(scale=12):
user_input = gr.Textbox(
show_label=False,
placeholder="Shift + Enter, to send message...",
lines=10).style(
container=False)
with gr.Column(min_width=32, scale=1):
submitBtn = gr.Button("Submit", variant="primary")
with gr.Column(scale=1):
emptyBtn = gr.Button("Clear History")
max_new_token = gr.Slider(
0,
4096,
value=512,
step=1.0,
label="Maximum New Token Length",
interactive=True)
top_p = gr.Slider(0, 1, value=0.9, step=0.01,
label="Top P", interactive=True)
temperature = gr.Slider(
0,
1,
value=0.5,
step=0.01,
label="Temperature",
interactive=True)
top_k = gr.Slider(1, 40, value=40, step=1,
label="Top K", interactive=True)
do_sample = gr.Checkbox(
value=True,
label="Do Sample",
info="use random sample strategy",
interactive=True)
repetition_penalty = gr.Slider(
1.0,
3.0,
value=1.1,
step=0.1,
label="Repetition Penalty",
interactive=True)
params = [user_input, chatbot]
predict_params = [
chatbot,
max_new_token,
top_p,
temperature,
top_k,
do_sample,
repetition_penalty]
submitBtn.click(
user,
params,
params,
queue=False).then(
predict,
predict_params,
chatbot).then(
lambda: gr.update(
interactive=True),
None,
[user_input],
queue=False)
user_input.submit(
user,
params,
params,
queue=False).then(
predict,
predict_params,
chatbot).then(
lambda: gr.update(
interactive=True),
None,
[user_input],
queue=False)
submitBtn.click(reset_user_input, [], [user_input])
emptyBtn.click(reset_state, outputs=[chatbot], show_progress=True)
# Launch the Gradio interface
demo.queue().launch()
"""
demo.queue().launch(
share=share,
inbrowser=True,
server_name='0.0.0.0',
server_port=port)
"""
"""
demo.queue().launch(
root_path="/etc/nginx/sites-available/radllama2_gradio_app"
)
"""
|