File size: 2,981 Bytes
5c9adbd
7138f42
 
 
 
 
5c9adbd
7138f42
 
5bfa3aa
 
7138f42
1cb615a
5c9adbd
 
 
 
 
 
 
 
 
7138f42
 
 
 
 
7d41378
5c9adbd
 
7138f42
d4772a6
7138f42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1cb615a
7138f42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2a78417
7138f42
 
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
import os
from threading import Thread
from typing import Iterator

import torch
from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from huggingface_hub import login

# model_id = 'meta-llama/Llama-2-13b-chat-hf'
#model_id = 'meta-llama/Llama-2-7b-chat-hf'
model_id = 'Trelis/Llama-2-7b-chat-hf-sharded-bf16'

if torch.cuda.is_available():
    tok = os.environ['HF_TOKEN']
    login(new_session=True,
          write_permission=False,
          token=tok
          
          #, token="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"
          )
    
    config = AutoConfig.from_pretrained(model_id, use_auth_token=True)
    config.pretraining_tp = 1
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        config=config,
        torch_dtype=torch.float16,
        #load_in_4bit=True,
        device_map='auto',
        use_auth_token=True
    )
    print("Loaded the model!")
else:
    model = None
tokenizer = AutoTokenizer.from_pretrained(model_id)


def get_prompt(message: str, chat_history: list[tuple[str, str]],
               system_prompt: str) -> str:
    texts = [f'<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n']
    # The first user input is _not_ stripped
    do_strip = False
    for user_input, response in chat_history:
        user_input = user_input.strip() if do_strip else user_input
        do_strip = True
        texts.append(f'{user_input} [/INST] {response.strip()} </s><s>[INST] ')
    message = message.strip() if do_strip else message
    texts.append(f'{message} [/INST]')
    return ''.join(texts)


def get_input_token_length(message: str, chat_history: list[tuple[str, str]], system_prompt: str) -> int:
    prompt = get_prompt(message, chat_history, system_prompt)
    input_ids = tokenizer([prompt], return_tensors='np', add_special_tokens=False)['input_ids']
    return input_ids.shape[-1]


def run(message: str,
        chat_history: list[tuple[str, str]],
        system_prompt: str,
        max_new_tokens: int = 1024,
        temperature: float = 0.8,
        top_p: float = 0.95,
        top_k: int = 50) -> Iterator[str]:
    prompt = get_prompt(message, chat_history, system_prompt)
    inputs = tokenizer([prompt], return_tensors='pt', add_special_tokens=False).to('cuda') #.to(torch.device) #

    streamer = TextIteratorStreamer(tokenizer,
                                    timeout=10.,
                                    skip_prompt=True,
                                    skip_special_tokens=True)
    generate_kwargs = dict(
        inputs,
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        print(f"output text ->{text}<- end of text")
        outputs.append(text)
        yield ''.join(outputs)