File size: 3,668 Bytes
6b2a98f
 
 
 
 
 
 
 
26bc57d
 
6b2a98f
76d3296
6b2a98f
 
 
 
 
 
 
 
 
 
 
 
 
 
26bc57d
6b2a98f
26bc57d
 
6b2a98f
 
 
 
 
 
 
 
 
 
 
 
 
e674021
6b2a98f
 
 
 
 
 
 
 
 
 
 
 
 
 
76d3296
 
6b2a98f
 
 
 
 
76d3296
6b2a98f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b93b962
 
6b2a98f
 
 
 
80a2cf1
b93b962
6b2a98f
 
 
 
 
 
 
 
 
fcf72a5
6b2a98f
 
 
 
 
 
 
 
26bc57d
6b2a98f
 
 
16cf867
6b2a98f
 
 
 
26bc57d
 
 
 
6b2a98f
 
 
26bc57d
6b2a98f
 
 
 
 
 
b93b962
76d3296
26bc57d
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
import gradio as gr
import huggingface_hub
from fastrtc import (
    AdditionalOutputs,
    ReplyOnPause,
    WebRTC,
    WebRTCData,
    WebRTCError,
    get_current_context,
    get_hf_turn_credentials,
    get_stt_model,
)
from gradio.utils import get_space
from openai import OpenAI

stt_model = get_stt_model()

conversations = {}


def response(
    data: WebRTCData,
    conversation: list[dict],
    model: str = "meta-llama/Llama-3.2-3B-Instruct",
    provider: str = "sambanova",
):
    context = get_current_context()
    print("conversation before", conversation)
    if not provider.startswith("http") and not context.oauth_token:
        raise WebRTCError("Please Sign in to use this demo.")

    if data.audio is not None and data.audio[1].size > 0:
        user_audio_text = stt_model.stt(data.audio)
        conversation.append({"role": "user", "content": user_audio_text})
    else:
        conversation.append({"role": "user", "content": data.textbox})

    yield AdditionalOutputs(conversation)

    if provider.startswith("http"):
        client = OpenAI(base_url=provider, api_key="ollama")
    else:
        client = huggingface_hub.InferenceClient(
            api_key=context.oauth_token.token,  # type: ignore
            provider=provider,  # type: ignore
        )

    request = client.chat.completions.create(
        model=model,
        messages=conversation,  # type: ignore
        temperature=1,
        top_p=0.1,
    )
    response = {"role": "assistant", "content": request.choices[0].message.content}

    conversation.append(response)
    print("conversation after", conversation)
    yield AdditionalOutputs(conversation)


css = """
footer {
    display: none !important;
}
"""

providers = [
    "black-forest-labs",
    "cerebras",
    "cohere",
    "fal-ai",
    "fireworks-ai",
    "hf-inference",
    "hyperbolic",
    "nebius",
    "novita",
    "openai",
    "replicate",
    "sambanova",
    "together",
]


def hide_token(provider: str):
    if provider.startswith("http"):
        return gr.Textbox(visible=False)
    return gr.skip()


with gr.Blocks(css=css) as demo:
    gr.HTML(
        """
            <h1 style='text-align: center; display: flex; align-items: center; justify-content: center;'>
            <img src="https://huggingface.co/datasets/freddyaboulton/bucket/resolve/main/AV_Huggy.png" alt="Streaming Huggy" style="height: 50px; margin-right: 10px"> FastRTC Chat
            </h1>
        """
    )
    with gr.Sidebar():
        gr.LoginButton()
        model = gr.Dropdown(
            choices=["meta-llama/Llama-3.2-3B-Instruct"],
            allow_custom_value=True,
            label="Model",
        )
        provider = gr.Dropdown(
            label="Provider",
            choices=providers,
            value="auto",
            info="Select a hf-compatible provider or type the url of your server, e.g. http://127.0.0.1:11434/v1 for ollama",
            allow_custom_value=True,
        )
    cb = gr.Chatbot(type="messages", height=400)
    webrtc = WebRTC(
        modality="audio",
        mode="send",
        variant="textbox",
        rtc_configuration=get_hf_turn_credentials if get_space() else None,
        server_rtc_configuration=get_hf_turn_credentials(ttl=3_600 * 24 * 30)
        if get_space()
        else None,
    )
    webrtc.stream(
        ReplyOnPause(response),  # type: ignore
        inputs=[webrtc, cb, model, provider],
        outputs=[cb],
        concurrency_limit=100,
    )
    webrtc.on_additional_outputs(
        lambda old, new: new, inputs=[cb], outputs=[cb], concurrency_limit=100
    )

if __name__ == "__main__":
    demo.launch(server_port=7860)