File size: 7,279 Bytes
b585c7f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import sys

import pytest

from src.utils import get_list_or_str, read_popen_pipes, get_token_count, reverse_ucurve_list, undo_reverse_ucurve_list
from tests.utils import wrap_test_forked
import subprocess as sp


@wrap_test_forked
def test_get_list_or_str():
    assert get_list_or_str(['foo', 'bar']) == ['foo', 'bar']
    assert get_list_or_str('foo') == 'foo'
    assert get_list_or_str("['foo', 'bar']") == ['foo', 'bar']


@wrap_test_forked
def test_stream_popen1():
    cmd_python = sys.executable + " -i -q -u"
    cmd = cmd_python + " -c print('hi')"
    # cmd = cmd.split(' ')

    with sp.Popen(cmd, stdout=sp.PIPE, stderr=sp.PIPE, text=True, shell=True) as p:
        for out_line, err_line in read_popen_pipes(p):
            print(out_line, end='')
            print(err_line, end='')

        p.poll()


@wrap_test_forked
def test_stream_popen2():
    script = """for i in 0 1 2 3 4 5
do
    echo "This messages goes to stdout $i"
    sleep 1
    echo This message goes to stderr >&2
    sleep 1
done
"""
    with open('pieces.sh', 'wt') as f:
        f.write(script)
    with sp.Popen(["./pieces.sh"], stdout=sp.PIPE, stderr=sp.PIPE, text=True, shell=True) as p:
        for out_line, err_line in read_popen_pipes(p):
            print(out_line, end='')
            print(err_line, end='')
        p.poll()


@pytest.mark.parametrize("text_context_list",
                         ['text_context_list1', 'text_context_list2', 'text_context_list3', 'text_context_list4',
                          'text_context_list5', 'text_context_list6'])
@pytest.mark.parametrize("system_prompt", ['auto', ''])
@pytest.mark.parametrize("context", ['context1', 'context2'])
@pytest.mark.parametrize("iinput", ['iinput1', 'iinput2'])
@pytest.mark.parametrize("chat_conversation", ['chat_conversation1', 'chat_conversation2'])
@pytest.mark.parametrize("instruction", ['instruction1', 'instruction2'])
@wrap_test_forked
def test_limited_prompt(instruction, chat_conversation, iinput, context, system_prompt, text_context_list):
    instruction1 = 'Who are you?'
    instruction2 = ' '.join(['foo_%s ' % x for x in range(0, 500)])
    instruction = instruction1 if instruction == 'instruction1' else instruction2

    iinput1 = 'Extra instruction info'
    iinput2 = ' '.join(['iinput_%s ' % x for x in range(0, 500)])
    iinput = iinput1 if iinput == 'iinput1' else iinput2

    context1 = 'context'
    context2 = ' '.join(['context_%s ' % x for x in range(0, 500)])
    context = context1 if context == 'context1' else context2

    chat_conversation1 = []
    chat_conversation2 = [['user_conv_%s ' % x, 'bot_conv_%s ' % x] for x in range(0, 500)]
    chat_conversation = chat_conversation1 if chat_conversation == 'chat_conversation1' else chat_conversation2

    text_context_list1 = []
    text_context_list2 = ['doc_%s ' % x for x in range(0, 500)]
    text_context_list3 = ['doc_%s ' % x for x in range(0, 10)]
    text_context_list4 = ['documentmany_%s ' % x for x in range(0, 10000)]
    import random, string
    text_context_list5 = [
        'documentlong_%s_%s' % (x, ''.join(random.choices(string.ascii_letters + string.digits, k=300))) for x in
        range(0, 20)]
    text_context_list6 = [
        'documentlong_%s_%s' % (x, ''.join(random.choices(string.ascii_letters + string.digits, k=4000))) for x in
        range(0, 1)]
    if text_context_list == 'text_context_list1':
        text_context_list = text_context_list1
    elif text_context_list == 'text_context_list2':
        text_context_list = text_context_list2
    elif text_context_list == 'text_context_list3':
        text_context_list = text_context_list3
    elif text_context_list == 'text_context_list4':
        text_context_list = text_context_list4
    elif text_context_list == 'text_context_list5':
        text_context_list = text_context_list5
    elif text_context_list == 'text_context_list6':
        text_context_list = text_context_list6
    else:
        raise ValueError("No such %s" % text_context_list)

    from transformers import AutoTokenizer
    tokenizer = AutoTokenizer.from_pretrained('h2oai/h2ogpt-4096-llama2-7b-chat')

    prompt_type = 'llama2'
    prompt_dict = None
    debug = False
    chat = True
    stream_output = True
    from src.prompter import Prompter
    prompter = Prompter(prompt_type, prompt_dict, debug=debug,
                        stream_output=stream_output,
                        system_prompt=system_prompt)

    min_max_new_tokens = 256  # like in get_limited_prompt()
    max_input_tokens = -1
    max_new_tokens = 1024
    model_max_length = 4096

    from src.gen import get_limited_prompt
    estimated_full_prompt, \
        instruction, iinput, context, \
        num_prompt_tokens, max_new_tokens, \
        num_prompt_tokens0, num_prompt_tokens_actual, \
        history_to_use_final, external_handle_chat_conversation, \
        top_k_docs_trial, one_doc_size, truncation_generation, system_prompt = \
        get_limited_prompt(instruction, iinput, tokenizer,
                           prompter=prompter,
                           max_new_tokens=max_new_tokens,
                           context=context,
                           chat_conversation=chat_conversation,
                           text_context_list=text_context_list,
                           model_max_length=model_max_length,
                           min_max_new_tokens=min_max_new_tokens,
                           max_input_tokens=max_input_tokens,
                           verbose=True)
    print('%s -> %s or %s: len(history_to_use_final): %s top_k_docs_trial=%s one_doc_size: %s' % (num_prompt_tokens0,
                                                                                   num_prompt_tokens,
                                                                                   num_prompt_tokens_actual,
                                                                                   len(history_to_use_final),
                                                                                   top_k_docs_trial,
                                                                                   one_doc_size),
          flush=True, file=sys.stderr)
    assert num_prompt_tokens <= model_max_length + min_max_new_tokens
    # actual might be less due to token merging for characters across parts, but not more
    assert num_prompt_tokens >= num_prompt_tokens_actual
    assert num_prompt_tokens_actual <= model_max_length

    if top_k_docs_trial > 0:
        text_context_list = text_context_list[:top_k_docs_trial]
    elif one_doc_size is not None:
        text_context_list = [text_context_list[0][:one_doc_size]]
    else:
        text_context_list = []
    assert sum([get_token_count(x, tokenizer) for x in text_context_list]) <= model_max_length


@wrap_test_forked
def test_reverse_ucurve():
    ab = []
    a = [1, 2, 3, 4, 5, 6, 7, 8]
    b = [2, 4, 6, 8, 7, 5, 3, 1]
    ab.append([a, b])
    a = [1]
    b = [1]
    ab.append([a, b])
    a = [1, 2]
    b = [2, 1]
    ab.append([a, b])
    a = [1, 2, 3]
    b = [2, 3, 1]
    ab.append([a, b])
    a = [1, 2, 3, 4]
    b = [2, 4, 3, 1]
    ab.append([a, b])

    for a, b in ab:
        assert reverse_ucurve_list(a) == b
        assert undo_reverse_ucurve_list(b) == a