Spaces:
Runtime error
Runtime error
File size: 12,070 Bytes
8acb22e |
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 |
# -*- coding:utf-8 -*-
from __future__ import annotations
from typing import TYPE_CHECKING, Any, Callable, Dict, List, Tuple, Type
import logging
# import json
# import os
# import datetime
# import hashlib
# import csv
# import requests
import re
import html
# import markdown2
import torch
import sys
import gc
from pygments.lexers import guess_lexer, ClassNotFound
import gradio as gr
# from pypinyin import lazy_pinyin
# import tiktoken
# import mdtex2html
# from markdown import markdown
from pygments import highlight
from pygments.lexers import guess_lexer,get_lexer_by_name
from pygments.formatters import HtmlFormatter
# import transformers
# from peft import PeftModel
# from transformers import GenerationConfig, LlamaForCausalLM, LlamaTokenizer
from app_modules.presets import *
# logging.basicConfig(
# level=logging.INFO,
# format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s",
# )
def markdown_to_html_with_syntax_highlight(md_str):
def replacer(match):
lang = match.group(1) or "text"
code = match.group(2)
lang = lang.strip()
#print(1,lang)
if lang=="text":
lexer = guess_lexer(code)
lang = lexer.name
#print(2,lang)
try:
lexer = get_lexer_by_name(lang, stripall=True)
except ValueError:
lexer = get_lexer_by_name("python", stripall=True)
formatter = HtmlFormatter()
#print(3,lexer.name)
highlighted_code = highlight(code, lexer, formatter)
return f'<pre><code class="{lang}">{highlighted_code}</code></pre>'
code_block_pattern = r"```(\w+)?\n([\s\S]+?)\n```"
md_str = re.sub(code_block_pattern, replacer, md_str, flags=re.MULTILINE)
html_str = markdown(md_str)
return html_str
def normalize_markdown(md_text: str) -> str:
lines = md_text.split("\n")
normalized_lines = []
inside_list = False
for i, line in enumerate(lines):
if re.match(r"^(\d+\.|-|\*|\+)\s", line.strip()):
if not inside_list and i > 0 and lines[i - 1].strip() != "":
normalized_lines.append("")
inside_list = True
normalized_lines.append(line)
elif inside_list and line.strip() == "":
if i < len(lines) - 1 and not re.match(
r"^(\d+\.|-|\*|\+)\s", lines[i + 1].strip()
):
normalized_lines.append(line)
continue
else:
inside_list = False
normalized_lines.append(line)
return "\n".join(normalized_lines)
def convert_mdtext(md_text):
code_block_pattern = re.compile(r"```(.*?)(?:```|$)", re.DOTALL)
inline_code_pattern = re.compile(r"`(.*?)`", re.DOTALL)
code_blocks = code_block_pattern.findall(md_text)
non_code_parts = code_block_pattern.split(md_text)[::2]
result = []
for non_code, code in zip(non_code_parts, code_blocks + [""]):
if non_code.strip():
non_code = normalize_markdown(non_code)
if inline_code_pattern.search(non_code):
result.append(markdown(non_code, extensions=["tables"]))
else:
result.append(mdtex2html.convert(non_code, extensions=["tables"]))
if code.strip():
code = f"\n```{code}\n\n```"
code = markdown_to_html_with_syntax_highlight(code)
result.append(code)
result = "".join(result)
result += ALREADY_CONVERTED_MARK
return result
def convert_asis(userinput):
return f"<p style=\"white-space:pre-wrap;\">{html.escape(userinput)}</p>"+ALREADY_CONVERTED_MARK
def detect_converted_mark(userinput):
if userinput.endswith(ALREADY_CONVERTED_MARK):
return True
else:
return False
def detect_language(code):
if code.startswith("\n"):
first_line = ""
else:
first_line = code.strip().split("\n", 1)[0]
language = first_line.lower() if first_line else ""
code_without_language = code[len(first_line) :].lstrip() if first_line else code
return language, code_without_language
def convert_to_markdown(text):
text = text.replace("$","$")
def replace_leading_tabs_and_spaces(line):
new_line = []
for char in line:
if char == "\t":
new_line.append("	")
elif char == " ":
new_line.append(" ")
else:
break
return "".join(new_line) + line[len(new_line):]
markdown_text = ""
lines = text.split("\n")
in_code_block = False
for line in lines:
if in_code_block is False and line.startswith("```"):
in_code_block = True
markdown_text += f"{line}\n"
elif in_code_block is True and line.startswith("```"):
in_code_block = False
markdown_text += f"{line}\n"
elif in_code_block:
markdown_text += f"{line}\n"
else:
line = replace_leading_tabs_and_spaces(line)
line = re.sub(r"^(#)", r"\\\1", line)
markdown_text += f"{line} \n"
return markdown_text
def add_language_tag(text):
def detect_language(code_block):
try:
lexer = guess_lexer(code_block)
return lexer.name.lower()
except ClassNotFound:
return ""
code_block_pattern = re.compile(r"(```)(\w*\n[^`]+```)", re.MULTILINE)
def replacement(match):
code_block = match.group(2)
if match.group(2).startswith("\n"):
language = detect_language(code_block)
if language:
return f"```{language}{code_block}```"
else:
return f"```\n{code_block}```"
else:
return match.group(1) + code_block + "```"
text2 = code_block_pattern.sub(replacement, text)
return text2
def delete_last_conversation(chatbot, history):
if len(chatbot) > 0:
chatbot.pop()
if len(history) > 0:
history.pop()
return (
chatbot,
history,
"Delete Done",
)
def reset_state():
return [], [], "Reset Done"
def reset_textbox():
return gr.update(value=""),""
def cancel_outputing():
return "Stop Done"
def transfer_input(inputs):
textbox = reset_textbox()
return (
inputs,
gr.update(value=""),
gr.Button.update(visible=True),
)
class State:
interrupted = False
def interrupt(self):
self.interrupted = True
def recover(self):
self.interrupted = False
shared_state = State()
# Greedy Search
def greedy_search(input_ids: torch.Tensor,
model: torch.nn.Module,
tokenizer: transformers.PreTrainedTokenizer,
stop_words: list,
max_length: int,
temperature: float = 1.0,
top_p: float = 1.0,
top_k: int = 25) -> Iterator[str]:
generated_tokens = []
past_key_values = None
current_length = 1
for i in range(max_length):
with torch.no_grad():
if past_key_values is None:
outputs = model(input_ids)
else:
outputs = model(input_ids[:, -1:], past_key_values=past_key_values)
logits = outputs.logits[:, -1, :]
past_key_values = outputs.past_key_values
# apply temperature
logits /= temperature
probs = torch.softmax(logits, dim=-1)
# apply top_p
probs_sort, probs_idx = torch.sort(probs, dim=-1, descending=True)
probs_sum = torch.cumsum(probs_sort, dim=-1)
mask = probs_sum - probs_sort > top_p
probs_sort[mask] = 0.0
# apply top_k
#if top_k is not None:
# probs_sort1, _ = torch.topk(probs_sort, top_k)
# min_top_probs_sort = torch.min(probs_sort1, dim=-1, keepdim=True).values
# probs_sort = torch.where(probs_sort < min_top_probs_sort, torch.full_like(probs_sort, float(0.0)), probs_sort)
probs_sort.div_(probs_sort.sum(dim=-1, keepdim=True))
next_token = torch.multinomial(probs_sort, num_samples=1)
next_token = torch.gather(probs_idx, -1, next_token)
input_ids = torch.cat((input_ids, next_token), dim=-1)
generated_tokens.append(next_token[0].item())
text = tokenizer.decode(generated_tokens)
yield text
if any([x in text for x in stop_words]):
del past_key_values
del logits
del probs
del probs_sort
del probs_idx
del probs_sum
gc.collect()
return
def generate_prompt_with_history(text,history,tokenizer,max_length=2048):
prompt = "The following is a conversation between a human and an AI assistant named Baize (named after a mythical creature in Chinese folklore). Baize is an open-source AI assistant developed by UCSD and Sun Yat-Sen University. The human and the AI assistant take turns chatting. Human statements start with [|Human|] and AI assistant statements start with [|AI|]. The AI assistant always provides responses in as much detail as possible, and in Markdown format. The AI assistant always declines to engage with topics, questions and instructions related to unethical, controversial, or sensitive issues. Complete the transcript in exactly that format.\n[|Human|]Hello!\n[|AI|]Hi!"
history = ["\n[|Human|]{}\n[|AI|]{}".format(x[0],x[1]) for x in history]
history.append("\n[|Human|]{}\n[|AI|]".format(text))
history_text = ""
flag = False
for x in history[::-1]:
if tokenizer(prompt+history_text+x, return_tensors="pt")['input_ids'].size(-1) <= max_length:
history_text = x + history_text
flag = True
else:
break
if flag:
return prompt+history_text,tokenizer(prompt+history_text, return_tensors="pt")
else:
return None
def is_stop_word_or_prefix(s: str, stop_words: list) -> bool:
for stop_word in stop_words:
if s.endswith(stop_word):
return True
for i in range(1, len(stop_word)):
if s.endswith(stop_word[:i]):
return True
return False
def load_tokenizer_and_model(base_model,adapter_model=None,load_8bit=False):
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
try:
if torch.backends.mps.is_available():
device = "mps"
except: # noqa: E722
pass
tokenizer = LlamaTokenizer.from_pretrained(base_model)
if device == "cuda":
model = LlamaForCausalLM.from_pretrained(
base_model,
load_in_8bit=load_8bit,
torch_dtype=torch.float16,
device_map="auto",
)
if adapter_model is not None:
model = PeftModel.from_pretrained(
model,
adapter_model,
torch_dtype=torch.float16,
)
elif device == "mps":
model = LlamaForCausalLM.from_pretrained(
base_model,
device_map={"": device},
torch_dtype=torch.float16,
)
if adapter_model is not None:
model = PeftModel.from_pretrained(
model,
adapter_model,
device_map={"": device},
torch_dtype=torch.float16,
)
else:
model = LlamaForCausalLM.from_pretrained(
base_model, device_map={"": device}, low_cpu_mem_usage=True
)
if adapter_model is not None:
model = PeftModel.from_pretrained(
model,
adapter_model,
device_map={"": device},
)
if not load_8bit:
model.half() # seems to fix bugs for some users.
model.eval()
return tokenizer,model,device
|