Spaces:
Sleeping
Sleeping
File size: 5,330 Bytes
b2b40b3 2360578 aec1ff0 5cda5a4 2360578 c11d9db 3d2c1c1 2360578 aec1ff0 2360578 aec1ff0 2360578 aec1ff0 2360578 aec1ff0 2360578 aec1ff0 2360578 aec1ff0 2360578 b259230 |
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 gradio as gr
import time
from transformers import Qwen2AudioForConditionalGeneration, AutoProcessor
from io import BytesIO
from urllib.request import urlopen
import librosa
import os, json
from sys import argv
from vllm import LLM, SamplingParams
from huggingface_hub import login
TOKEN = os.environ.get("TOKEN", None)
login(token=TOKEN)
def load_model_processor(model_path):
processor = AutoProcessor.from_pretrained(model_path)
llm = LLM(
model=model_path, trust_remote_code=True, gpu_memory_utilization=0.8,
enforce_eager=True, device = "cuda",
limit_mm_per_prompt={"audio": 5},
)
return llm, processor
model_path1 = "SeaLLMs/SeaLLMs-Audio-7B"
model1, processor1 = load_model_processor(model_path1)
def response_to_audio(audio_url, text, model=None, processor=None, temperature = 0.1,repetition_penalty=1.1, top_p = 0.9,max_new_tokens = 2048):
if text == None:
conversation = [
{"role": "user", "content": [
{"type": "audio", "audio_url": audio_url},
]},]
elif audio_url == None:
conversation = [
{"role": "user", "content": [
{"type": "text", "text": text},
]},]
else:
conversation = [
{"role": "user", "content": [
{"type": "audio", "audio_url": audio_url},
{"type": "text", "text": text},
]},]
text = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
audios = []
for message in conversation:
if isinstance(message["content"], list):
for ele in message["content"]:
if ele["type"] == "audio":
if ele['audio_url'] != None:
audios.append(librosa.load(
ele['audio_url'],
sr=processor.feature_extractor.sampling_rate)[0]
)
sampling_params = SamplingParams(
temperature=temperature, max_tokens=max_new_tokens, repetition_penalty=repetition_penalty, top_p=top_p, top_k=20,
stop_token_ids=[],
)
input = {
'prompt': text,
'multi_modal_data': {
'audio': [(audio, 16000) for audio in audios]
}
}
output = model.generate([input], sampling_params=sampling_params)[0]
response = output.outputs[0].text
return response
def clear_inputs():
return None, "", ""
def compare_responses(audio_url, text):
response1 = response_to_audio(audio_url, text, model1, processor1)
return response1
with gr.Blocks() as demo:
# gr.Markdown(f"Evaluate {model_path1}")
# gr.Markdown("""<p align="center"><img src="images/seal_logo.png" style="height: 80px"/><p>""")
# gr.Image("images/seal_logo.png", elem_id="seal_logo", show_label=False,height=80,show_fullscreen_button=False)
gr.Markdown(
"""<div style="text-align: center; font-size: 32px; font-weight: bold;">SeaLLMs-Audio ChatBot</div>""",
)
# Description text
gr.Markdown(
"""<div style="text-align: center; font-size: 16px;">
This WebUI is based on SeaLLMs-Audio-7B-Chat, developed by Alibaba DAMO Academy.<br>
You can interact with the chatbot in <b>English, Chinese, Indonesian, Thai, or Vietnamese</b>.<br>
For each round, you can input <b>audio and/or text</b>.
</div>""",
)
# Links with proper formatting
gr.Markdown(
"""<div style="text-align: center; font-size: 16px;">
<a href="https://huggingface.co/SeaLLMs/SeaLLMs-v3-7B-Chat">[Website]</a>
<a href="https://huggingface.co/SeaLLMs/SeaLLMs-v3-7B-Chat">[Model🤗]</a>
<a href="https://github.com/liuchaoqun/SeaLLMs-Audio">[Github]</a>
</div>""",
)
# gr.Markdown(insturctions)
# with gr.Row():
# with gr.Column():
# temperature = gr.Slider(minimum=0, maximum=1, value=0.3, step=0.1, label="Temperature")
# with gr.Column():
# top_p = gr.Slider(minimum=0.1, maximum=1, value=0.5, step=0.1, label="Top P")
# with gr.Column():
# repetition_penalty = gr.Slider(minimum=0, maximum=2, value=1.1, step=0.1, label="Repetition Penalty")
with gr.Row():
with gr.Column():
# mic_input = gr.Microphone(label="Record Audio", type="filepath", elem_id="mic_input")
mic_input = gr.Audio(sources = ['upload', 'microphone'], label="Record Audio", type="filepath", elem_id="mic_input")
with gr.Column():
additional_input = gr.Textbox(label="Text Input")
# Button to trigger the function
with gr.Row():
btn_submit = gr.Button("Submit")
btn_clear = gr.Button("Clear")
with gr.Row():
output_text1 = gr.Textbox(label=model_path1.split('/')[-1], interactive=False, elem_id="output_text1")
btn_submit.click(
fn=response_to_audio,
inputs=[mic_input, additional_input],
outputs=[output_text1],
)
btn_clear.click(
fn=clear_inputs,
inputs=None,
outputs=[mic_input, additional_input, output_text1],
queue=False,
)
demo.launch(
share=False,
inbrowser=True,
server_port=7950,
server_name="0.0.0.0",
max_threads=40
)
|