File size: 4,947 Bytes
9f341cc
 
 
 
 
 
 
 
 
 
e916990
 
 
 
 
6aa8b86
9f341cc
 
 
6aa8b86
 
 
 
 
e820e51
9f341cc
 
 
 
 
 
 
 
d2b20f2
9f341cc
 
 
6aa8b86
2da6968
9f341cc
e820e51
 
9f341cc
 
 
 
 
 
2da6968
 
 
 
 
 
 
395ee29
 
 
 
 
e820e51
9f341cc
 
 
3dfcb72
9f341cc
 
 
d2b20f2
9f341cc
b96cef7
9f341cc
d2b20f2
9f341cc
 
 
 
d2b20f2
9f341cc
d2b20f2
047008b
 
 
 
 
d2b20f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9f341cc
 
 
 
 
 
 
395ee29
d2b20f2
9f341cc
 
047008b
9f341cc
d2b20f2
 
 
 
 
 
 
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
import json
import re
import requests
from messagers.message_outputer import OpenaiStreamOutputer
from utils.logger import logger
from utils.enver import enver


class MessageStreamer:
    MODEL_MAP = {
        "mixtral-8x7b": "mistralai/Mixtral-8x7B-Instruct-v0.1",  # 72.62, fast [Recommended]
        "mistral-7b": "mistralai/Mistral-7B-Instruct-v0.2",  # 65.71, fast
        "openchat-3.5": "openchat/openchat_3.5",  # 61.24, fast
        # "zephyr-7b-alpha": "HuggingFaceH4/zephyr-7b-alpha",  # 59.5, fast
        # "zephyr-7b-beta": "HuggingFaceH4/zephyr-7b-beta",  # 61.95, slow
        "default": "mistralai/Mixtral-8x7B-Instruct-v0.1",
    }

    def __init__(self, model: str):
        if model in self.MODEL_MAP.keys():
            self.model = model
        else:
            self.model = "default"
        self.model_fullname = self.MODEL_MAP[self.model]
        self.message_outputer = OpenaiStreamOutputer()

    def parse_line(self, line):
        line = line.decode("utf-8")
        line = re.sub(r"data:\s*", "", line)
        data = json.loads(line)
        content = data["token"]["text"]
        return content

    def chat_response(
        self,
        prompt: str = None,
        temperature: float = 0.01,
        max_new_tokens: int = 8192,
        api_key: str = None,
    ):
        # https://huggingface.co/docs/api-inference/detailed_parameters?code=curl
        # 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>'
        self.request_url = (
            f"https://api-inference.huggingface.co/models/{self.model_fullname}"
        )
        self.request_headers = {
            "Content-Type": "application/json",
        }

        if api_key:
            logger.note(
                f"Using API Key: {api_key[:3]}{(len(api_key)-7)*'*'}{api_key[-4:]}"
            )
            self.request_headers["Authorization"] = f"Bearer {api_key}"

        # References:
        #   huggingface_hub/inference/_client.py:
        #     class InferenceClient > def text_generation()
        #   huggingface_hub/inference/_text_generation.py:
        #     class TextGenerationRequest > param `stream`
        # https://huggingface.co/docs/text-generation-inference/conceptual/streaming#streaming-with-curl
        self.request_body = {
            "inputs": prompt,
            "parameters": {
                "temperature": max(temperature, 0.01),  # must be positive
                "max_new_tokens": max_new_tokens,
                "return_full_text": False,
            },
            "stream": True,
        }
        logger.back(self.request_url)
        enver.set_envs(proxies=True)
        stream_response = requests.post(
            self.request_url,
            headers=self.request_headers,
            json=self.request_body,
            proxies=enver.requests_proxies,
            stream=True,
        )
        status_code = stream_response.status_code
        if status_code == 200:
            logger.success(status_code)
        else:
            logger.err(status_code)

        return stream_response

    def chat_return_dict(self, stream_response):
        # https://platform.openai.com/docs/guides/text-generation/chat-completions-response-format
        final_output = self.message_outputer.default_data.copy()
        final_output["choices"] = [
            {
                "index": 0,
                "finish_reason": "stop",
                "message": {
                    "role": "assistant",
                    "content": "",
                },
            }
        ]
        logger.back(final_output)

        for line in stream_response.iter_lines():
            if not line:
                continue
            content = self.parse_line(line)

            if content.strip() == "</s>":
                logger.success("\n[Finished]")
                break
            else:
                logger.back(content, end="")
                final_output["choices"][0]["message"]["content"] += content

        return final_output

    def chat_return_generator(self, stream_response):
        is_finished = False
        for line in stream_response.iter_lines():
            if not line:
                continue

            content = self.parse_line(line)

            if content.strip() == "</s>":
                content_type = "Finished"
                logger.success("\n[Finished]")
                is_finished = True
            else:
                content_type = "Completions"
                logger.back(content, end="")

            output = self.message_outputer.output(
                content=content, content_type=content_type
            )
            yield output

        if not is_finished:
            yield self.message_outputer.output(content="", content_type="Finished")