feat: update app
Browse files- app.py +164 -51
- assets/assistant_avavar.png +0 -0
- requirements.txt +3 -1
app.py
CHANGED
@@ -1,64 +1,177 @@
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
-
from huggingface_hub import InferenceClient
|
3 |
|
4 |
-
|
5 |
-
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
|
6 |
-
"""
|
7 |
-
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
8 |
|
9 |
|
10 |
-
|
11 |
-
|
12 |
-
history: list[tuple[str, str]],
|
13 |
-
system_message,
|
14 |
-
max_tokens,
|
15 |
-
temperature,
|
16 |
-
top_p,
|
17 |
-
):
|
18 |
-
messages = [{"role": "system", "content": system_message}]
|
19 |
|
20 |
-
|
21 |
-
if val[0]:
|
22 |
-
messages.append({"role": "user", "content": val[0]})
|
23 |
-
if val[1]:
|
24 |
-
messages.append({"role": "assistant", "content": val[1]})
|
25 |
|
26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
stream=True,
|
|
|
|
|
34 |
temperature=temperature,
|
35 |
top_p=top_p,
|
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 |
-
if __name__ ==
|
64 |
-
|
|
|
1 |
+
import gc
|
2 |
+
import os
|
3 |
+
|
4 |
import gradio as gr
|
|
|
5 |
|
6 |
+
from llama_cpp import Llama
|
|
|
|
|
|
|
7 |
|
8 |
|
9 |
+
ALPACA_SYSTEM_PROMPT = 'Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request'
|
10 |
+
ALPACA_SYSTEM_PROMPT_NO_INPUT = 'Below is an instruction that describes a task. Write a response that appropriately completes the request.'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
+
DEFAULT_MODEL = 'Med-Alpaca-2-7b-chat.Q4_K_M'
|
|
|
|
|
|
|
|
|
13 |
|
14 |
+
model_paths = {
|
15 |
+
'Med-Alpaca-2-7b-chat.Q2_K': {
|
16 |
+
'repo_id': 'minhnguyent546/Med-Alpaca-2-7b-chat-GGUF',
|
17 |
+
'filename': 'Med-Alpaca-2-7B-chat.Q2_K.gguf',
|
18 |
+
},
|
19 |
+
'Med-Alpaca-2-7b-chat.Q4_K_M': {
|
20 |
+
'repo_id': 'minhnguyent546/Med-Alpaca-2-7b-chat-GGUF',
|
21 |
+
'filename': 'Med-Alpaca-2-7B-chat.Q4_K_M.gguf',
|
22 |
+
},
|
23 |
+
'Med-Alpaca-2-7b-chat.Q6_K': {
|
24 |
+
'repo_id': 'minhnguyent546/Med-Alpaca-2-7b-chat-GGUF',
|
25 |
+
'filename': 'Med-Alpaca-2-7B-chat.Q6_K.gguf',
|
26 |
+
},
|
27 |
+
'Med-Alpaca-2-7b-chat.Q8_0': {
|
28 |
+
'repo_id': 'minhnguyent546/Med-Alpaca-2-7b-chat-GGUF',
|
29 |
+
'filename': 'Med-Alpaca-2-7B-chat.Q8_0.gguf',
|
30 |
+
},
|
31 |
+
'Med-Alpaca-2-7b-chat.F16': {
|
32 |
+
'repo_id': 'minhnguyent546/Med-Alpaca-2-7b-chat-GGUF',
|
33 |
+
'filename': 'Med-Alpaca-2-7B-chat.F16.gguf',
|
34 |
+
},
|
35 |
+
}
|
36 |
+
|
37 |
+
model = Llama.from_pretrained(
|
38 |
+
**model_paths[DEFAULT_MODEL],
|
39 |
+
n_ctx=4096,
|
40 |
+
n_threads=4,
|
41 |
+
cache_dir='./hf-cache'
|
42 |
+
)
|
43 |
|
44 |
+
def generate_alpaca_prompt(
|
45 |
+
instruction: str,
|
46 |
+
input: str | None = None,
|
47 |
+
response: str = '',
|
48 |
+
) -> str:
|
49 |
+
prompt = ''
|
50 |
+
if input is not None and input and input.strip() != '<noinput>':
|
51 |
+
prompt = (
|
52 |
+
f'{ALPACA_SYSTEM_PROMPT}\n\n'
|
53 |
+
f'### Instruction:\n'
|
54 |
+
f'{instruction}\n\n'
|
55 |
+
f'### Input:\n'
|
56 |
+
f'{input}\n\n'
|
57 |
+
f'### Response: '
|
58 |
+
f'{response}'
|
59 |
+
)
|
60 |
+
else:
|
61 |
+
prompt = (
|
62 |
+
f'{ALPACA_SYSTEM_PROMPT_NO_INPUT}\n\n'
|
63 |
+
f'### Instruction:\n'
|
64 |
+
f'{instruction}\n\n'
|
65 |
+
f'### Response: '
|
66 |
+
f'{response}'
|
67 |
+
)
|
68 |
+
return prompt.strip()
|
69 |
|
70 |
+
def chat_completion(
|
71 |
+
message,
|
72 |
+
history,
|
73 |
+
seed: int,
|
74 |
+
max_new_tokens: int,
|
75 |
+
temperature: float,
|
76 |
+
repeatition_penalty: float,
|
77 |
+
top_k: int,
|
78 |
+
top_p: float,
|
79 |
+
):
|
80 |
+
prompt = generate_alpaca_prompt(instruction=message)
|
81 |
+
response_iterator = model(
|
82 |
+
prompt,
|
83 |
stream=True,
|
84 |
+
seed=seed,
|
85 |
+
max_tokens=max_new_tokens,
|
86 |
temperature=temperature,
|
87 |
top_p=top_p,
|
88 |
+
top_k=top_k,
|
89 |
+
repeat_penalty=repeatition_penalty,
|
90 |
+
)
|
91 |
+
partial_response = ''
|
92 |
+
for token in response_iterator:
|
93 |
+
partial_response += token['choices'][0]['text']
|
94 |
+
yield partial_response
|
95 |
+
|
96 |
+
def on_model_changed(model_name: str):
|
97 |
+
global model
|
98 |
+
if 'model' in globals():
|
99 |
+
del model
|
100 |
+
gc.collect()
|
101 |
+
model = Llama.from_pretrained(
|
102 |
+
**model_paths[model_name],
|
103 |
+
n_ctx=4096,
|
104 |
+
n_threads=4,
|
105 |
+
cache_dir='./hf-cache'
|
106 |
+
)
|
107 |
+
|
108 |
+
app_title_mark = gr.Markdown(f"""<center><font size=16>{model_name}</center>""")
|
109 |
+
chatbot = gr.Chatbot(
|
110 |
+
type='messages',
|
111 |
+
height=500,
|
112 |
+
placeholder='<strong>Hi, I have a headache, what should I do?</strong>',
|
113 |
+
label=model_name,
|
114 |
+
avatar_images=[None, './assets/assistant_avavar.png'], # pyright: ignore[reportArgumentType]
|
115 |
+
)
|
116 |
+
return app_title_mark, chatbot
|
117 |
+
|
118 |
+
def main() -> None:
|
119 |
+
with gr.Blocks(theme=gr.themes.Ocean()) as demo:
|
120 |
+
app_title_mark = gr.Markdown(f"""<center><font size=18>{DEFAULT_MODEL}</center>""")
|
121 |
+
|
122 |
+
model_options = list(model_paths.keys())
|
123 |
+
|
124 |
+
with gr.Row():
|
125 |
+
with gr.Column(scale=2):
|
126 |
+
with gr.Row():
|
127 |
+
model_radio = gr.Radio(choices=model_options, label='Model', value=DEFAULT_MODEL)
|
128 |
+
with gr.Row():
|
129 |
+
seed = gr.Number(value=998244353, label='Seed')
|
130 |
+
max_new_tokens = gr.Number(value=512, minimum=64, maximum=2048, label='Max new tokens')
|
131 |
+
|
132 |
+
with gr.Row():
|
133 |
+
temperature = gr.Slider(0, 2, step=0.01, label='Temperature', value=0.6, info='Info')
|
134 |
+
repeatition_penalty = gr.Slider(0.01, 5, step=0.05, label='Repetition penalty', value=1.1)
|
135 |
+
|
136 |
+
with gr.Row():
|
137 |
+
top_k = gr.Slider(1, 100, step=1, label='Top k', value=40)
|
138 |
+
top_p = gr.Slider(0, 1, step=0.01, label='Top p', value=0.9)
|
139 |
+
|
140 |
+
with gr.Column(scale=5):
|
141 |
+
chatbot = gr.Chatbot(
|
142 |
+
type='messages',
|
143 |
+
height=500,
|
144 |
+
placeholder='<strong>Hi, I have a headache, what should I do?</strong>',
|
145 |
+
label=DEFAULT_MODEL,
|
146 |
+
avatar_images=[None, './assets/assistant_avavar.png'], # pyright: ignore[reportArgumentType]
|
147 |
+
)
|
148 |
+
textbox = gr.Textbox(
|
149 |
+
placeholder='Hi, I have a headache, what should I do?',
|
150 |
+
container=False,
|
151 |
+
submit_btn=True,
|
152 |
+
stop_btn=True,
|
153 |
+
)
|
154 |
+
|
155 |
+
chat_interface = gr.ChatInterface(
|
156 |
+
chat_completion,
|
157 |
+
type='messages',
|
158 |
+
chatbot=chatbot,
|
159 |
+
textbox=textbox,
|
160 |
+
additional_inputs=[
|
161 |
+
seed,
|
162 |
+
max_new_tokens,
|
163 |
+
temperature,
|
164 |
+
repeatition_penalty,
|
165 |
+
top_k,
|
166 |
+
top_p,
|
167 |
+
],
|
168 |
+
)
|
169 |
+
|
170 |
+
model_radio.change(on_model_changed, inputs=[model_radio], outputs=[app_title_mark, chatbot])
|
171 |
+
|
172 |
+
demo.queue(api_open=False, default_concurrency_limit=20)
|
173 |
+
demo.launch(max_threads=5, share=os.environ.get('GRADIO_SHARE', False))
|
174 |
|
175 |
|
176 |
+
if __name__ == '__main__':
|
177 |
+
main()
|
assets/assistant_avavar.png
ADDED
![]() |
requirements.txt
CHANGED
@@ -1 +1,3 @@
|
|
1 |
-
|
|
|
|
|
|
1 |
+
gradio~=5.6.0
|
2 |
+
huggingface_hub==0.25.2
|
3 |
+
llama-cpp-python~=0.3.2
|