FazBrowse GitHub Viewer
|
Trending
|
URL:
|
Home
Tools:
[Download Repo ZIP]
[View Raw Code]
[Original HTTPS Page]
Medusa/data_generation/generate.py at main · FasterDecoding/Medusa · GitHub
Uh oh!
There was an error while loading.
Please reload this page
.
FasterDecoding
/
Medusa
Public
Notifications
You must be signed in to change notification settings
Fork
205
Star
2.8k
Code
Issues
51
Pull requests
6
Discussions
Actions
Projects
Security and quality
0
Insights
Additional navigation options
Code
Issues
Pull requests
Discussions
Actions
Projects
Security and quality
Insights
Expand file tree
Breadcrumbs
Medusa
/
data_generation
/
generate.py
Copy path
More file actions
More file actions
Latest commit
History
History
History
156 lines (140 loc) · 5.16 KB
Breadcrumbs
Medusa
/
data_generation
/
generate.py
Copy path
File metadata and controls
156 lines (140 loc) · 5.16 KB
Raw
Copy raw file
Download raw file
Open symbols panel
Edit and raw actions
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
import
json
import
os
import
time
import
concurrent
.
futures
import
openai
import
shortuuid
import
tqdm
import
argparse
import
random
from
tenacity
import
(
retry
,
stop_after_attempt
,
wait_random_exponential
,
)
from
fastchat
.
conversation
import
Conversation
,
SeparatorStyle
from
fastchat
.
model
.
model_adapter
import
get_conversation_template
# Modify OpenAI's API key and API base to use vLLM's API server.
openai
.
api_key
=
"EMPTY"
openai
.
api_base
=
"http://localhost:8000/v1"
api_base_pool
=
[]
# List models API
for
i
in
range
(
10
):
openai
.
api_base
=
"http://localhost:800{}/v1"
.
format
(
i
)
try
:
models
=
openai
.
Model
.
list
()[
"data"
][
0
][
"id"
]
print
(
openai
.
api_base
,
models
)
api_base_pool
.
append
(
openai
.
api_base
)
except
:
break
print
(
"API base pool: "
,
api_base_pool
)
parser
=
argparse
.
ArgumentParser
()
parser
.
add_argument
(
"--data_path"
,
type
=
str
)
parser
.
add_argument
(
"--output_path"
,
type
=
str
)
parser
.
add_argument
(
"--num_threads"
,
type
=
int
,
default
=
256
)
parser
.
add_argument
(
"--temperature"
,
type
=
float
,
default
=
0.3
)
parser
.
add_argument
(
"--max_tokens"
,
type
=
int
,
default
=
2048
)
parser
.
add_argument
(
"--chat"
,
action
=
"store_true"
)
args
=
parser
.
parse_args
()
# Assuming the ShareGPT format
data
=
json
.
load
(
open
(
args
.
data_path
,
"r"
))
def
generate_data
(
messages
,
idx
):
try
:
# load balanced
openai
.
api_base
=
api_base_pool
[
idx
%
len
(
api_base_pool
)]
model_name
=
openai
.
Model
.
list
()[
"data"
][
0
][
"id"
]
if
args
.
chat
:
converted_messages
=
[]
output_messages
=
[]
if
messages
[
0
][
"from"
]
==
"system"
:
converted_messages
.
append
(
{
"role"
:
"system"
,
"content"
:
messages
[
0
][
"text"
],
}
)
output_messages
.
append
(
messages
[
0
])
messages
=
messages
[
1
:]
for
message
in
messages
[::
2
]:
if
message
[
"from"
]
!=
"human"
:
return
converted_messages
.
append
(
{
"role"
:
"user"
,
"content"
:
message
[
"value"
],
}
)
try
:
response
=
openai
.
ChatCompletion
.
create
(
model
=
model_name
,
messages
=
converted_messages
,
max_tokens
=
args
.
max_tokens
,
temperature
=
args
.
temperature
,
)
if
response
.
choices
[
0
][
'finish_reason'
]
==
"length"
:
break
response
=
response
.
choices
[
0
][
'message'
][
'content'
].
strip
()
output_messages
.
append
(
message
)
output_messages
.
append
(
{
"from"
:
"gpt"
,
"value"
:
response
,
}
)
converted_messages
.
append
(
{
"role"
:
"assistant"
,
"content"
:
response
,
}
)
except
:
break
if
len
(
output_messages
)
==
0
:
return
with
open
(
args
.
output_path
,
"a"
)
as
f
:
# write in share gpt format
f
.
write
(
json
.
dumps
({
"conversations"
:
output_messages
})
+
"
\n
"
)
else
:
conv
=
get_conversation_template
(
model_name
)
if
messages
[
0
][
"from"
]
==
"system"
:
conv
.
system_message
=
messages
[
0
][
"text"
]
messages
=
messages
[
1
:]
conv
.
append_message
(
conv
.
roles
[
0
],
messages
[
0
][
"value"
])
conv
.
append_message
(
conv
.
roles
[
1
],
None
)
prompt
=
conv
.
get_prompt
()
response
=
openai
.
Completion
.
create
(
model
=
model_name
,
prompt
=
prompt
,
max_tokens
=
args
.
max_tokens
,
temperature
=
args
.
temperature
,
ignore_eos
=
True
,
skip_special_tokens
=
False
,
spaces_between_special_tokens
=
False
,
)
response
=
response
.
choices
[
0
][
'text'
].
strip
()
with
open
(
args
.
output_path
,
"a"
)
as
f
:
# write in share gpt format
f
.
write
(
json
.
dumps
({
"text"
:
prompt
+
response
})
+
"
\n
"
)
except
Exception
as
e
:
print
(
e
)
print
(
prompt
)
print
(
"Failed to generate data"
)
# if output_path exists, count the number of lines and skip the first n data
start
=
0
if
os
.
path
.
exists
(
args
.
output_path
):
with
open
(
args
.
output_path
,
"r"
)
as
f
:
start
=
len
(
f
.
readlines
())
print
(
"Skip first {} data"
.
format
(
start
))
with
concurrent
.
futures
.
ThreadPoolExecutor
(
max_workers
=
args
.
num_threads
)
as
executor
:
futures
=
[]
for
idx
,
sample
in
enumerate
(
data
[
start
:]):
future
=
executor
.
submit
(
generate_data
,
sample
[
"conversations"
],
idx
,
)
futures
.
append
(
future
)
for
future
in
tqdm
.
tqdm
(
concurrent
.
futures
.
as_completed
(
futures
),
total
=
len
(
futures
)
):
future
.
result
()
Back
|
FazBrowse Home
|
New Git URL