This repository has been archived by the owner on Mar 30, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 17
/
streamers.py
229 lines (218 loc) · 7.23 KB
/
streamers.py
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
import json
from os import times
from ctransformers import LLM, Config
from log import log
from get_env import get_env
from const import DEFAULT_CONTEXT_LENGTH, DEFAULT_LOG_LEVEL
def completions_streamer(
prompt: str,
model_name: str,
llm: LLM,
config: Config,
):
"""_summary_
returns a generator that yields a stream of responses
"""
created = times()
top_k = config.top_k
log.debug("top_k: %s", top_k)
top_p = config.top_p
log.debug("top_p: %s", top_p)
temperature = config.temperature
log.debug("temperature: %s", temperature)
repetition_penalty = config.repetition_penalty
log.debug("repetition_penalty: %s", repetition_penalty)
last_n_tokens = config.last_n_tokens
log.debug("last_n_tokens: %s", last_n_tokens)
seed = config.seed
log.debug("seed: %s", seed)
batch_size = config.batch_size
log.debug("batch_size: %s", batch_size)
threads = config.threads
log.debug("threads: %s", threads)
max_new_tokens = config.max_new_tokens
log.debug("max_new_tokens: %s", max_new_tokens)
stop = config.stop
log.debug("stop: %s", stop)
log.debug("prompt: %s", prompt)
CONTEXT_LENGTH = int(get_env("CONTEXT_LENGTH", DEFAULT_CONTEXT_LENGTH))
LOGGING_LEVEL = get_env("LOGGING_LEVEL", DEFAULT_LOG_LEVEL)
log.debug("Streaming from ctransformer instance!")
total_tokens = 0
for token in llm(
prompt,
stream=True,
reset=True,
top_k=top_k,
top_p=top_p,
temperature=temperature,
repetition_penalty=repetition_penalty,
last_n_tokens=last_n_tokens,
seed=seed,
batch_size=batch_size,
threads=threads,
max_new_tokens=max_new_tokens,
stop=stop,
):
if LOGGING_LEVEL == "DEBUG":
# Only track token length if we're in debug mode to avoid overhead
total_tokens = total_tokens + len(token)
# tokens are not necessarily characters, but this is a good enough approximation
if total_tokens > CONTEXT_LENGTH:
log.debug(
"Total token length %s exceeded context length %s",
total_tokens,
CONTEXT_LENGTH,
)
log.debug(
"Try to increase CONTEXT_LENGTH that is currently set to %s to your model's context length",
CONTEXT_LENGTH,
)
log.debug(
"Alternatively, increse REPETITION_PENALTY %s and LAST_N_TOKENS %s AND/OR adjust temperature %s top_k %s top_p %s",
repetition_penalty,
last_n_tokens,
temperature,
top_k,
top_p,
)
log.debug("Streaming token %s", token)
data = json.dumps(
{
"id": "id",
"object": "text_completion.chunk",
"created": created,
"model": model_name,
"choices": [
{
"text": token,
"index": 0,
"finish_reason": None,
}
],
}
)
yield f"data: {data}" + "\n\n"
stop_data = json.dumps(
{
"id": "id",
"object": "text_completion.chunk",
"created": created,
"model": model_name,
"choices": [
{
"text": "",
"index": 0,
"finish_reason": "stop",
}
],
}
)
yield f"data: {stop_data}" + "\n\n"
log.debug("Streaming ended")
def chat_completions_streamer(
prompt: str,
model_name: str,
llm: LLM,
config: Config,
):
"""_summary_
returns a generator that yields a stream of responses
"""
created = times()
top_k = config.top_k
log.debug("top_k: %s", top_k)
top_p = config.top_p
log.debug("top_p: %s", top_p)
temperature = config.temperature
log.debug("temperature: %s", temperature)
repetition_penalty = config.repetition_penalty
log.debug("repetition_penalty: %s", repetition_penalty)
last_n_tokens = config.last_n_tokens
log.debug("last_n_tokens: %s", last_n_tokens)
seed = config.seed
log.debug("seed: %s", seed)
batch_size = config.batch_size
log.debug("batch_size: %s", batch_size)
threads = config.threads
log.debug("threads: %s", threads)
max_new_tokens = config.max_new_tokens
log.debug("max_new_tokens: %s", max_new_tokens)
stop = config.stop
log.debug("stop: %s", stop)
log.debug("prompt: %s", prompt)
CONTEXT_LENGTH = int(get_env("CONTEXT_LENGTH", DEFAULT_CONTEXT_LENGTH))
LOGGING_LEVEL = get_env("LOGGING_LEVEL", DEFAULT_LOG_LEVEL)
log.debug("Streaming from ctransformer instance")
total_tokens = 0
for token in llm(
prompt,
stream=True,
reset=True,
top_k=top_k,
top_p=top_p,
temperature=temperature,
repetition_penalty=repetition_penalty,
last_n_tokens=last_n_tokens,
seed=seed,
batch_size=batch_size,
threads=threads,
max_new_tokens=max_new_tokens,
stop=stop,
):
if LOGGING_LEVEL == "DEBUG":
# Only track token length if we're in debug mode to avoid overhead
total_tokens = total_tokens + len(token)
# tokens are not necessarily characters, but this is a good enough approximation
if total_tokens > CONTEXT_LENGTH:
log.debug(
"Total token length %s exceeded context length %s",
total_tokens,
CONTEXT_LENGTH,
)
log.debug(
"Try to increase CONTEXT_LENGTH that is currently set to %s to your model's context length",
CONTEXT_LENGTH,
)
log.debug(
"Alternatively, increse REPETITION_PENALTY %s and LAST_N_TOKENS %s AND/OR adjust temperature %s top_k %s top_p %s",
repetition_penalty,
last_n_tokens,
temperature,
top_k,
top_p,
)
log.debug("Streaming token %s", token)
data = json.dumps(
{
"id": "id",
"object": "chat.completion.chunk",
"created": created,
"model": model_name,
"choices": [
{
"delta": {"role": "assistant", "content": token},
"index": 0,
"finish_reason": None,
}
],
}
)
yield f"data: {data}" + "\n\n"
stop_data = json.dumps(
{
"id": "id",
"object": "chat.completion.chunk",
"created": created,
"model": model_name,
"choices": [
{
"delta": {},
"index": 0,
"finish_reason": "stop",
}
],
}
)
yield f"data: {stop_data}" + "\n\n"
log.debug("Streaming ended")