Spaces:
Runtime error
Runtime error
Commit
·
8e06205
1
Parent(s):
7bb24c5
Create model.py
Browse files
model.py
ADDED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from threading import Thread
|
2 |
+
from typing import Iterator
|
3 |
+
|
4 |
+
#import torch
|
5 |
+
from transformers.utils import logging
|
6 |
+
from ctransformers import AutoModelForCausalLM, AutoTokenizer
|
7 |
+
from transformers import TextIteratorStreamer
|
8 |
+
|
9 |
+
logging.set_verbosity_info()
|
10 |
+
logger = logging.get_logger("transformers")
|
11 |
+
|
12 |
+
config = {'max_new_tokens': 256, 'repetition_penalty': 1.1,
|
13 |
+
'temperature': 0.1, 'stream': True}
|
14 |
+
model_id = 'TheBloke/Llama-2-7B-Chat-GGML'
|
15 |
+
device = "cpu"
|
16 |
+
|
17 |
+
|
18 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, model_type="llama", lib='avx2', hf=True)
|
19 |
+
tokenizer = AutoTokenizer.from_pretrained(model)
|
20 |
+
|
21 |
+
def get_prompt(message: str, chat_history: list[tuple[str, str]],
|
22 |
+
system_prompt: str) -> str:
|
23 |
+
logger.info("get_prompt chat_history=%s",chat_history)
|
24 |
+
logger.info("get_prompt system_prompt=%s",system_prompt)
|
25 |
+
texts = [f'<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n']
|
26 |
+
logger.info("texts=%s",texts)
|
27 |
+
# The first user input is _not_ stripped
|
28 |
+
do_strip = False
|
29 |
+
for user_input, response in chat_history:
|
30 |
+
user_input = user_input.strip() if do_strip else user_input
|
31 |
+
do_strip = True
|
32 |
+
texts.append(f'{user_input} [/INST] {response.strip()} </s><s>[INST] ')
|
33 |
+
message = message.strip() if do_strip else message
|
34 |
+
logger.info("get_prompt message=%s",message)
|
35 |
+
texts.append(f'{message} [/INST]')
|
36 |
+
logger.info("get_prompt final texts=%s",texts)
|
37 |
+
return ''.join(texts)
|
38 |
+
|
39 |
+
|
40 |
+
def get_input_token_length(message: str, chat_history: list[tuple[str, str]], system_prompt: str) -> int:
|
41 |
+
logger.info("get_input_token_length=%s",message)
|
42 |
+
prompt = get_prompt(message, chat_history, system_prompt)
|
43 |
+
logger.info("prompt=%s",prompt)
|
44 |
+
input_ids = tokenizer([prompt], return_tensors='np', add_special_tokens=False)['input_ids']
|
45 |
+
logger.info("input_ids=%s",input_ids)
|
46 |
+
return input_ids.shape[-1]
|
47 |
+
|
48 |
+
|
49 |
+
def run(message: str,
|
50 |
+
chat_history: list[tuple[str, str]],
|
51 |
+
system_prompt: str,
|
52 |
+
max_new_tokens: int = 1024,
|
53 |
+
temperature: float = 0.8,
|
54 |
+
top_p: float = 0.95,
|
55 |
+
top_k: int = 50) -> Iterator[str]:
|
56 |
+
prompt = get_prompt(message, chat_history, system_prompt)
|
57 |
+
inputs = tokenizer([prompt], return_tensors='pt', add_special_tokens=False).to(device)
|
58 |
+
|
59 |
+
streamer = TextIteratorStreamer(tokenizer,
|
60 |
+
timeout=10.,
|
61 |
+
skip_prompt=True,
|
62 |
+
skip_special_tokens=True)
|
63 |
+
generate_kwargs = dict(
|
64 |
+
inputs,
|
65 |
+
streamer=streamer,
|
66 |
+
max_new_tokens=max_new_tokens,
|
67 |
+
do_sample=True,
|
68 |
+
top_p=top_p,
|
69 |
+
top_k=top_k,
|
70 |
+
temperature=temperature,
|
71 |
+
num_beams=1,
|
72 |
+
bits=4,
|
73 |
+
groupsize=128,
|
74 |
+
)
|
75 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
76 |
+
t.start()
|
77 |
+
|
78 |
+
outputs = []
|
79 |
+
for text in streamer:
|
80 |
+
outputs.append(text)
|
81 |
+
yield ''.join(outputs)
|