File size: 3,882 Bytes
ec0c335
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Call API providers."""

import os
import random
import time

from fastchat.utils import build_logger
from fastchat.constants import WORKER_API_TIMEOUT


logger = build_logger("gradio_web_server", "gradio_web_server.log")


def openai_api_stream_iter(
    model_name,
    messages,
    temperature,
    top_p,
    max_new_tokens,
    api_base=None,
    api_key=None,
):
    import openai

    is_azure = False
    if "azure" in model_name:
        is_azure = True
        openai.api_type = "azure"
        openai.api_version = "2023-07-01-preview"
    else:
        openai.api_type = "open_ai"
        openai.api_version = None

    openai.api_base = api_base or "https://api.openai.com/v1"
    openai.api_key = api_key or os.environ["OPENAI_API_KEY"]
    if model_name == "gpt-4-turbo":
        model_name = "gpt-4-1106-preview"

    # Make requests
    gen_params = {
        "model": model_name,
        "prompt": messages,
        "temperature": temperature,
        "top_p": top_p,
        "max_new_tokens": max_new_tokens,
    }
    logger.info(f"==== request ====\n{gen_params}")

    if is_azure:
        res = openai.ChatCompletion.create(
            engine=model_name,
            messages=messages,
            temperature=temperature,
            max_tokens=max_new_tokens,
            stream=True,
        )
    else:
        res = openai.ChatCompletion.create(
            model=model_name,
            messages=messages,
            temperature=temperature,
            max_tokens=max_new_tokens,
            stream=True,
        )
    text = ""
    for chunk in res:
        if len(chunk["choices"]) > 0:
            text += chunk["choices"][0]["delta"].get("content", "")
            data = {
                "text": text,
                "error_code": 0,
            }
            yield data


def anthropic_api_stream_iter(model_name, prompt, temperature, top_p, max_new_tokens):
    import anthropic

    c = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

    # Make requests
    gen_params = {
        "model": model_name,
        "prompt": prompt,
        "temperature": temperature,
        "top_p": top_p,
        "max_new_tokens": max_new_tokens,
    }
    logger.info(f"==== request ====\n{gen_params}")

    res = c.completions.create(
        prompt=prompt,
        stop_sequences=[anthropic.HUMAN_PROMPT],
        max_tokens_to_sample=max_new_tokens,
        temperature=temperature,
        top_p=top_p,
        model=model_name,
        stream=True,
    )
    text = ""
    for chunk in res:
        text += chunk.completion
        data = {
            "text": text,
            "error_code": 0,
        }
        yield data


def init_palm_chat(model_name):
    import vertexai  # pip3 install google-cloud-aiplatform
    from vertexai.preview.language_models import ChatModel

    project_id = os.environ["GCP_PROJECT_ID"]
    location = "us-central1"
    vertexai.init(project=project_id, location=location)

    chat_model = ChatModel.from_pretrained(model_name)
    chat = chat_model.start_chat(examples=[])
    return chat


def palm_api_stream_iter(chat, message, temperature, top_p, max_new_tokens):
    parameters = {
        "temperature": temperature,
        "top_p": top_p,
        "max_output_tokens": max_new_tokens,
    }
    gen_params = {
        "model": "palm-2",
        "prompt": message,
    }
    gen_params.update(parameters)
    logger.info(f"==== request ====\n{gen_params}")

    response = chat.send_message(message, **parameters)
    content = response.text

    pos = 0
    while pos < len(content):
        # This is a fancy way to simulate token generation latency combined
        # with a Poisson process.
        pos += random.randint(10, 20)
        time.sleep(random.expovariate(50))
        data = {
            "text": content[:pos],
            "error_code": 0,
        }
        yield data