:gem: [Feature] Support no-stream mode with dict response
Browse files- apis/chat_api.py +17 -12
- messagers/message_outputer.py +10 -8
- mocks/stream_chat_mocker.py +4 -3
- networks/message_streamer.py +47 -13
apis/chat_api.py
CHANGED
@@ -64,19 +64,24 @@ class ChatAPIApp:
|
|
64 |
streamer = MessageStreamer(model=item.model)
|
65 |
composer = MessageComposer(model=item.model)
|
66 |
composer.merge(messages=item.messages)
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
yield_output=True,
|
74 |
-
),
|
75 |
-
media_type="text/event-stream",
|
76 |
-
ping=2000,
|
77 |
-
ping_message_factory=lambda: ServerSentEvent(**{"comment": ""}),
|
78 |
)
|
79 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
|
81 |
def setup_routes(self):
|
82 |
for prefix in ["", "/v1"]:
|
|
|
64 |
streamer = MessageStreamer(model=item.model)
|
65 |
composer = MessageComposer(model=item.model)
|
66 |
composer.merge(messages=item.messages)
|
67 |
+
# streamer.chat = stream_chat_mock
|
68 |
+
|
69 |
+
stream_response = streamer.chat_response(
|
70 |
+
prompt=composer.merged_str,
|
71 |
+
temperature=item.temperature,
|
72 |
+
max_new_tokens=item.max_tokens,
|
|
|
|
|
|
|
|
|
|
|
73 |
)
|
74 |
+
if item.stream:
|
75 |
+
event_source_response = EventSourceResponse(
|
76 |
+
streamer.chat_return_generator(stream_response),
|
77 |
+
media_type="text/event-stream",
|
78 |
+
ping=2000,
|
79 |
+
ping_message_factory=lambda: ServerSentEvent(**{"comment": ""}),
|
80 |
+
)
|
81 |
+
return event_source_response
|
82 |
+
else:
|
83 |
+
data_response = streamer.chat_return_dict(stream_response)
|
84 |
+
return data_response
|
85 |
|
86 |
def setup_routes(self):
|
87 |
for prefix in ["", "/v1"]:
|
messagers/message_outputer.py
CHANGED
@@ -7,20 +7,22 @@ class OpenaiStreamOutputer:
|
|
7 |
* https://platform.openai.com/docs/api-reference/chat/create
|
8 |
"""
|
9 |
|
10 |
-
def
|
11 |
-
|
12 |
-
|
13 |
-
return data_str
|
14 |
-
|
15 |
-
def output(self, content=None, content_type="Completions") -> str:
|
16 |
-
data = {
|
17 |
"created": 1700000000,
|
18 |
"id": "chatcmpl-hugginface",
|
19 |
"object": "chat.completion.chunk",
|
20 |
-
# "content_type":
|
21 |
"model": "hugginface",
|
22 |
"choices": [],
|
23 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
if content_type == "Role":
|
25 |
data["choices"] = [
|
26 |
{
|
|
|
7 |
* https://platform.openai.com/docs/api-reference/chat/create
|
8 |
"""
|
9 |
|
10 |
+
def __init__(self):
|
11 |
+
self.default_data = {
|
|
|
|
|
|
|
|
|
|
|
12 |
"created": 1700000000,
|
13 |
"id": "chatcmpl-hugginface",
|
14 |
"object": "chat.completion.chunk",
|
15 |
+
# "content_type": "Completions",
|
16 |
"model": "hugginface",
|
17 |
"choices": [],
|
18 |
}
|
19 |
+
|
20 |
+
def data_to_string(self, data={}, content_type=""):
|
21 |
+
data_str = f"{json.dumps(data)}"
|
22 |
+
return data_str
|
23 |
+
|
24 |
+
def output(self, content=None, content_type="Completions") -> str:
|
25 |
+
data = self.default_data.copy()
|
26 |
if content_type == "Role":
|
27 |
data["choices"] = [
|
28 |
{
|
mocks/stream_chat_mocker.py
CHANGED
@@ -2,10 +2,11 @@ import time
|
|
2 |
from utils.logger import logger
|
3 |
|
4 |
|
5 |
-
def stream_chat_mock():
|
6 |
-
|
|
|
7 |
content = f"W{i+1} "
|
8 |
-
time.sleep(1
|
9 |
logger.mesg(content, end="")
|
10 |
yield content
|
11 |
logger.mesg("")
|
|
|
2 |
from utils.logger import logger
|
3 |
|
4 |
|
5 |
+
def stream_chat_mock(*args, **kwargs):
|
6 |
+
logger.note(msg=str(args) + str(kwargs))
|
7 |
+
for i in range(10):
|
8 |
content = f"W{i+1} "
|
9 |
+
time.sleep(0.1)
|
10 |
logger.mesg(content, end="")
|
11 |
yield content
|
12 |
logger.mesg("")
|
networks/message_streamer.py
CHANGED
@@ -31,13 +31,11 @@ class MessageStreamer:
|
|
31 |
content = data["token"]["text"]
|
32 |
return content
|
33 |
|
34 |
-
def
|
35 |
self,
|
36 |
prompt: str = None,
|
37 |
temperature: float = 0.01,
|
38 |
max_new_tokens: int = 8192,
|
39 |
-
stream: bool = True,
|
40 |
-
yield_output: bool = False,
|
41 |
):
|
42 |
# https://huggingface.co/docs/api-inference/detailed_parameters?code=curl
|
43 |
# curl --proxy http://<server>:<port> https://api-inference.huggingface.co/models/<org>/<model_name> -X POST -d '{"inputs":"who are you?","parameters":{"max_new_token":64}}' -H 'Content-Type: application/json' -H 'Authorization: Bearer <HF_TOKEN>'
|
@@ -60,24 +58,57 @@ class MessageStreamer:
|
|
60 |
"max_new_tokens": max_new_tokens,
|
61 |
"return_full_text": False,
|
62 |
},
|
63 |
-
"stream":
|
64 |
}
|
65 |
logger.back(self.request_url)
|
66 |
enver.set_envs(proxies=True)
|
67 |
-
|
68 |
self.request_url,
|
69 |
headers=self.request_headers,
|
70 |
json=self.request_body,
|
71 |
proxies=enver.requests_proxies,
|
72 |
-
stream=
|
73 |
)
|
74 |
-
status_code =
|
75 |
if status_code == 200:
|
76 |
logger.success(status_code)
|
77 |
else:
|
78 |
logger.err(status_code)
|
79 |
|
80 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
if not line:
|
82 |
continue
|
83 |
|
@@ -86,12 +117,15 @@ class MessageStreamer:
|
|
86 |
if content.strip() == "</s>":
|
87 |
content_type = "Finished"
|
88 |
logger.success("\n[Finished]")
|
|
|
89 |
else:
|
90 |
content_type = "Completions"
|
91 |
logger.back(content, end="")
|
92 |
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
|
|
|
|
|
31 |
content = data["token"]["text"]
|
32 |
return content
|
33 |
|
34 |
+
def chat_response(
|
35 |
self,
|
36 |
prompt: str = None,
|
37 |
temperature: float = 0.01,
|
38 |
max_new_tokens: int = 8192,
|
|
|
|
|
39 |
):
|
40 |
# https://huggingface.co/docs/api-inference/detailed_parameters?code=curl
|
41 |
# curl --proxy http://<server>:<port> https://api-inference.huggingface.co/models/<org>/<model_name> -X POST -d '{"inputs":"who are you?","parameters":{"max_new_token":64}}' -H 'Content-Type: application/json' -H 'Authorization: Bearer <HF_TOKEN>'
|
|
|
58 |
"max_new_tokens": max_new_tokens,
|
59 |
"return_full_text": False,
|
60 |
},
|
61 |
+
"stream": True,
|
62 |
}
|
63 |
logger.back(self.request_url)
|
64 |
enver.set_envs(proxies=True)
|
65 |
+
stream_response = requests.post(
|
66 |
self.request_url,
|
67 |
headers=self.request_headers,
|
68 |
json=self.request_body,
|
69 |
proxies=enver.requests_proxies,
|
70 |
+
stream=True,
|
71 |
)
|
72 |
+
status_code = stream_response.status_code
|
73 |
if status_code == 200:
|
74 |
logger.success(status_code)
|
75 |
else:
|
76 |
logger.err(status_code)
|
77 |
|
78 |
+
return stream_response
|
79 |
+
|
80 |
+
def chat_return_dict(self, stream_response):
|
81 |
+
# https://platform.openai.com/docs/guides/text-generation/chat-completions-response-format
|
82 |
+
final_output = self.message_outputer.default_data.copy()
|
83 |
+
final_output["choices"] = [
|
84 |
+
{
|
85 |
+
"index": 0,
|
86 |
+
"finish_reason": "stop",
|
87 |
+
"message": {
|
88 |
+
"role": "assistant",
|
89 |
+
"content": "",
|
90 |
+
},
|
91 |
+
}
|
92 |
+
]
|
93 |
+
logger.back(final_output)
|
94 |
+
|
95 |
+
for line in stream_response.iter_lines():
|
96 |
+
if not line:
|
97 |
+
continue
|
98 |
+
content = self.parse_line(line)
|
99 |
+
|
100 |
+
if content.strip() == "</s>":
|
101 |
+
logger.success("\n[Finished]")
|
102 |
+
break
|
103 |
+
else:
|
104 |
+
logger.back(content, end="")
|
105 |
+
final_output["choices"][0]["message"]["content"] += content
|
106 |
+
|
107 |
+
return final_output
|
108 |
+
|
109 |
+
def chat_return_generator(self, stream_response):
|
110 |
+
is_finished = False
|
111 |
+
for line in stream_response.iter_lines():
|
112 |
if not line:
|
113 |
continue
|
114 |
|
|
|
117 |
if content.strip() == "</s>":
|
118 |
content_type = "Finished"
|
119 |
logger.success("\n[Finished]")
|
120 |
+
is_finished = True
|
121 |
else:
|
122 |
content_type = "Completions"
|
123 |
logger.back(content, end="")
|
124 |
|
125 |
+
output = self.message_outputer.output(
|
126 |
+
content=content, content_type=content_type
|
127 |
+
)
|
128 |
+
yield output
|
129 |
+
|
130 |
+
if not is_finished:
|
131 |
+
yield self.message_outputer.output(content="", content_type="Finished")
|