File size: 6,109 Bytes
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
08ba387
cf29295
c0eba71
 
 
08ba387
 
 
 
caac1c2
 
 
 
 
08ba387
 
 
 
 
 
 
 
cf29295
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
caac1c2
 
 
 
 
 
 
 
 
 
 
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf29295
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
08ba387
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
 
8af42db
c0eba71
 
 
 
 
5d9d8e7
c0eba71
 
 
 
 
5d9d8e7
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44be4ab
 
 
 
 
 
6a3a6de
44be4ab
 
c0eba71
 
 
 
 
 
 
 
 
 
 
 
 
5d9d8e7
 
 
d9eb9c3
5d9d8e7
9de96b1
d9eb9c3
c0eba71
 
 
 
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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
import spaces
import json
import subprocess
from llama_cpp import Llama
from llama_cpp_agent import LlamaCppAgent
from llama_cpp_agent.providers import LlamaCppPythonProvider
from llama_cpp_agent.chat_history import BasicChatHistory
from llama_cpp_agent.chat_history.messages import Roles
import gradio as gr
from huggingface_hub import hf_hub_download

hf_hub_download(
    repo_id="PartAI/Dorna-Llama3-8B-Instruct-GGUF",
    filename="dorna-llama3-8b-instruct.Q4_0.gguf",
    local_dir = "."
)
hf_hub_download(
    repo_id="PartAI/Dorna-Llama3-8B-Instruct-GGUF",
    filename="dorna-llama3-8b-instruct.Q8_0.gguf",
    local_dir = "."
)

hf_hub_download(
    repo_id="PartAI/Dorna-Llama3-8B-Instruct-GGUF",
    filename="dorna-llama3-8b-instruct.Q4_0.gguf",
    local_dir = "."
)
hf_hub_download(
    repo_id="PartAI/Dorna-Llama3-8B-Instruct-GGUF",
    filename="dorna-llama3-8b-instruct.Q5_0.gguf",
    local_dir = "."
)
hf_hub_download(
    repo_id="PartAI/Dorna-Llama3-8B-Instruct-GGUF",
    filename="dorna-llama3-8b-instruct.bf16.gguf",
    local_dir = "."
)

css = """
.message-row {
    justify-content: space-evenly !important;
}
.message-bubble-border {
    border-radius: 6px !important;
}
.dark.message-bubble-border {
    border-color: #343140 !important;
}
.dark.user {
    background: #1e1c26 !important;
}
.dark.assistant.dark, .dark.pending.dark {
    background: #16141c !important;
}
@import url('https://fonts.googleapis.com/css2?family=Vazirmatn&display=swap');
body, .gradio-container, .gr-button, .gr-input, .gr-slider, .gr-dropdown, .gr-markdown {
    font-family: 'Vazirmatn', sans-serif !important;
}
._button {
    font-size: 20px;
}
pre, code {
    direction: ltr !important;
    unicode-bidi: plaintext !important;
}
"""

def get_messages_formatter_type(model_name):
    from llama_cpp_agent import MessagesFormatterType
    return MessagesFormatterType.CHATML

@spaces.GPU(duration=120)
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    top_k,
    repeat_penalty,
    model,
):
    chat_template = get_messages_formatter_type(model)

    llm = Llama(
        model_path=f"./{model}",
        flash_attn=True,
        n_threads=40,
        n_gpu_layers=81,
        n_batch=1024,
        n_ctx=8192,
    )
    provider = LlamaCppPythonProvider(llm)

    agent = LlamaCppAgent(
        provider,
        system_prompt=f"{system_message}",
        predefined_messages_formatter_type=chat_template,
        debug_output=True
    )
    
    settings = provider.get_provider_default_settings()
    settings.temperature = temperature
    settings.top_k = top_k
    settings.top_p = top_p
    settings.max_tokens = max_tokens
    settings.repeat_penalty = repeat_penalty
    settings.stream = True

    messages = BasicChatHistory()

    for msn in history:
        user = {
            'role': Roles.user,
            'content': msn[0]
        }
        assistant = {
            'role': Roles.assistant,
            'content': msn[1]
        }
        messages.add_message(user)
        messages.add_message(assistant)
    
    stream = agent.get_chat_response(
        message[-2:],
        llm_sampling_settings=settings,
        chat_history=messages,
        returns_streaming_generator=True,
        print_output=False
    )
    
    outputs = ""
    for output in stream:
        outputs += output
        yield outputs

PLACEHOLDER = """
<div class="message-bubble-border" style="display:flex; max-width: 600px; border-radius: 8px; box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1); backdrop-filter: blur(10px);">
    <div style="padding: .5rem 1.5rem;">
    <h2 dir="rtl" style="text-align: right; font-size: 1.5rem; font-weight: 700; margin-bottom: 0.5rem;">با فرمت‌های GGUF درنا چت کنید!</h2>
</div>
"""
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a helpful Persian assistant. Please answer questions in the asked language.", label="System message", rtl=False),
        gr.Slider(minimum=1, maximum=8192, value=2048, step=1, label="Max tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.85,
            step=0.05,
            label="Top-p",
        ),
        gr.Slider(
            minimum=0,
            maximum=100,
            value=40,
            step=1,
            label="Top-k",
        ),
        gr.Slider(
            minimum=0.0,
            maximum=2.0,
            value=1,
            step=0.1,
            label="Repetition penalty",
        ),
        gr.Dropdown([
                'dorna-llama3-8b-instruct.Q8_0.gguf',
                'dorna-llama3-8b-instruct.Q4_0.gguf',
                'dorna-llama3-8b-instruct.Q5_0.gguf',
                'dorna-llama3-8b-instruct.bf16.gguf',
            ],
            value="dorna-llama3-8b-instruct.Q4_0.gguf",
            label="Model"
        ),
    ],
    theme=gr.themes.Soft(primary_hue="violet", secondary_hue="violet", neutral_hue="gray",font=[gr.themes.GoogleFont("Exo"), "ui-sans-serif", "system-ui", "sans-serif"]).set(
        body_background_fill_dark="#16141c",
        block_background_fill_dark="#16141c",
        block_border_width="1px",
        block_title_background_fill_dark="#1e1c26",
        input_background_fill_dark="#292733",
        button_secondary_background_fill_dark="#24212b",
        border_color_primary_dark="#343140",
        background_fill_secondary_dark="#16141c",
        color_accent_soft_dark="transparent"
    ),
    css=css,
    retry_btn="🔄 تلاش مجدد",
    undo_btn="↩️ بازگشت",
    clear_btn="🗑️ پاک کردن",
    submit_btn="ارسال",
    title="درنا، محصول مرکز تحقیقات هوش مصنوعی پارت",
    textbox=gr.Textbox(show_label=False, lines=2, rtl=True, placeholder="ورودی", show_copy_button=True, scale=4),
    chatbot=gr.Chatbot(scale=1, placeholder=PLACEHOLDER, rtl=True)
)

if __name__ == "__main__":
    demo.launch()