Hugging Face's logo

Spaces:
bardsai
/
performance-llm-board
Running

App Files Community
performance-llm-board
File size: 4,597 Bytes 
0b07a42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import os
from dataclasses import dataclass

import pandas as pd


@dataclass
class Model(object):
    model_display_name: str
    model_name: str
    api_url: str
    provider: str
    hourly_cost: int = None
    cost: str = None
    supports_functions: str = False
    size_billion_parameters: int = None  # in billion paramters
    cost_per_million_tokens: int = None
    cost_per_million_input_tokens: int = None
    cost_per_million_output_tokens: int = None

    def __post_init__(self):
        self.cost_per_million_input_tokens = self.cost_per_million_input_tokens or self.cost_per_million_tokens
        self.cost_per_million_output_tokens = self.cost_per_million_output_tokens or self.cost_per_million_tokens
        if not self.cost and self.hourly_cost:
            self.cost = f"${self.hourly_cost} / hour"
        if not self.cost and self.cost_per_million_tokens:
            self.cost = f"${self.cost_per_million_tokens} / 1M tokens"
        elif not self.cost and self.cost_per_million_input_tokens and self.cost_per_million_output_tokens:
            self.cost = f"${self.cost_per_million_input_tokens} / 1M input tokens, ${self.cost_per_million_output_tokens} / 1M output tokens"


env = os.environ

MODELS = [
    # source: https://openai.com/pricing
    # converted costs from dollar/1K tokens to dollar/1M for readability and together_ai comparability
    Model(
        "gpt-3.5-turbo",
        "gpt-3.5-turbo",
        None,
        "OpenAI",
        supports_functions=True,
        cost_per_million_input_tokens=1,
        cost_per_million_output_tokens=2,
    ),
    Model(
        "gpt-4-turbo",
        "gpt-4-1106-preview",
        None,
        "OpenAI",
        supports_functions=True,
        cost_per_million_input_tokens=10,
        cost_per_million_output_tokens=30,
    ),
    Model(
        "gpt-4",
        "gpt-4",
        None,
        "OpenAI",
        supports_functions=True,
        cost_per_million_input_tokens=30,
        cost_per_million_output_tokens=60,
    ),
    # we don't test gpt-4-32k because the tasks don't reach gpt-4 limitations
    Model(
        "gpt-3.5-turbo",
        "gpt-3.5-turbo",
        None,
        "OpenAI",
        supports_functions=True,
        cost_per_million_input_tokens=1,
        cost_per_million_output_tokens=2,
    ),
    # source: https://www.together.ai/pricing
    Model(
        "llama-2-70b-chat",
        "together_ai/togethercomputer/llama-2-70b-chat",
        None,
        "Together AI",
        cost_per_million_tokens=0.2,
    ),
    Model(
        "Mixtral-8x7B-Instruct-v0.1",
        "together_ai/mistralai/Mixtral-8x7B-Instruct-v0.1",
        None,
        "Together AI",
        size_billion_parameters=8 * 7,
        cost_per_million_tokens=0.9,
    ),
    # taken from endpoint pages
    Model(
        "zephyr-7b-beta",
        "huggingface/HuggingFaceH4/zephyr-7b-beta",
        env["ZEPHYR_7B_BETA_URL"],
        "Hugging Face Inference Endpoint",
        hourly_cost=1.30,
        size_billion_parameters=7,
    ),
    Model(
        "Mistral-7B-Instruct-v0.2",
        "huggingface/mistralai/Mistral-7B-Instruct-v0.2",
        env["MISTRAL_7B_BETA_URL"],
        "Hugging Face Inference Endpoint",
        hourly_cost=1.30,
        size_billion_parameters=7,
    ),
    Model(
        "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
        "huggingface/TinyLlama/TinyLlama-1.1B-Chat-v1.0",
        env["TINY_LLAMA_URL"],
        "Hugging Face Inference Endpoint",
        hourly_cost=0.60,
        size_billion_parameters=1.1,
    ),
    Model(
        "gemini-pro",
        "gemini-pro",
        None,
        "Google VertexAI",
        # https://ai.google.dev/pricing
        cost="$0.25 / 1M input characters, $0.5 / 1K output characters (60 queries per minute are free)",
        cost_per_million_input_tokens=0.25,
        cost_per_million_output_tokens=0.5,
    ),
    Model(
        "chat-bison",
        "chat-bison",
        None,
        "Google VertexAI",
        # https://cloud.google.com/vertex-ai/docs/generative-ai/pricing
        cost_per_million_input_tokens=0.25,
        cost_per_million_output_tokens=0.5,
    ),
    Model(
        "chat-bison-32k",
        "chat-bison-32k",
        None,
        "Google VertexAI",
        # https://cloud.google.com/vertex-ai/docs/generative-ai/pricing
        cost_per_million_input_tokens=0.25,
        cost_per_million_output_tokens=0.5,
    ),
]


def models_costs():
    return pd.DataFrame(
        [(model.model_display_name, model.provider, model.cost) for model in MODELS],
        columns=["Model", "Provider", "Cost"],
    )