Commit
•
4195e71
1
Parent(s):
f7a0c83
Update app.py
Browse files
app.py
CHANGED
@@ -2,16 +2,17 @@ from huggingface_hub import InferenceClient
|
|
2 |
import gradio as gr
|
3 |
from deep_translator import GoogleTranslator
|
4 |
|
|
|
5 |
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
|
6 |
|
7 |
-
#
|
8 |
-
def translate_to_english(text):
|
9 |
-
return GoogleTranslator(source='arabic', target='english').translate(text)
|
10 |
-
|
11 |
-
# Function to translate English text to Arabic
|
12 |
def translate_to_arabic(text):
|
13 |
-
return GoogleTranslator(source='
|
|
|
|
|
|
|
14 |
|
|
|
15 |
def format_prompt(message, history):
|
16 |
prompt = "<s>"
|
17 |
for user_prompt, bot_response in history:
|
@@ -20,88 +21,50 @@ def format_prompt(message, history):
|
|
20 |
prompt += f"[INST] {message} [/INST]"
|
21 |
return prompt
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
repetition_penalty=repetition_penalty,
|
40 |
-
do_sample=True,
|
41 |
-
seed=42,
|
42 |
-
)
|
43 |
-
|
44 |
-
# Translate the Arabic prompt to English
|
45 |
-
english_prompt = translate_to_english(prompt)
|
46 |
-
|
47 |
-
formatted_prompt = format_prompt(english_prompt, history)
|
48 |
-
|
49 |
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
|
|
50 |
output = ""
|
51 |
-
|
52 |
for response in stream:
|
53 |
-
output += response
|
54 |
|
55 |
# Translate the English response back to Arabic
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
gr.Slider(
|
65 |
-
label="Temperature",
|
66 |
-
value=0.9,
|
67 |
-
minimum=0.0,
|
68 |
-
maximum=1.0,
|
69 |
-
step=0.05,
|
70 |
-
interactive=True,
|
71 |
-
info="Higher values produce more diverse outputs",
|
72 |
-
),
|
73 |
-
gr.Slider(
|
74 |
-
label="Max new tokens",
|
75 |
-
value=256,
|
76 |
-
minimum=0,
|
77 |
-
maximum=1048,
|
78 |
-
step=64,
|
79 |
-
interactive=True,
|
80 |
-
info="The maximum numbers of new tokens",
|
81 |
-
),
|
82 |
-
gr.Slider(
|
83 |
-
label="Top-p (nucleus sampling)",
|
84 |
-
value=0.90,
|
85 |
-
minimum=0.0,
|
86 |
-
maximum=1,
|
87 |
-
step=0.05,
|
88 |
-
interactive=True,
|
89 |
-
info="Higher values sample more low-probability tokens",
|
90 |
-
),
|
91 |
-
gr.Slider(
|
92 |
-
label="Repetition penalty",
|
93 |
-
value=1.2,
|
94 |
-
minimum=1.0,
|
95 |
-
maximum=2.0,
|
96 |
-
step=0.05,
|
97 |
-
interactive=True,
|
98 |
-
info="Penalize repeated tokens",
|
99 |
-
)
|
100 |
]
|
101 |
|
102 |
-
|
|
|
103 |
fn=generate,
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import gradio as gr
|
3 |
from deep_translator import GoogleTranslator
|
4 |
|
5 |
+
# Initialize the Hugging Face client with your model
|
6 |
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
|
7 |
|
8 |
+
# Define the translation functions
|
|
|
|
|
|
|
|
|
9 |
def translate_to_arabic(text):
|
10 |
+
return GoogleTranslator(source='auto', target='ar').translate(text)
|
11 |
+
|
12 |
+
def translate_to_english(text):
|
13 |
+
return GoogleTranslator(source='auto', target='en').translate(text)
|
14 |
|
15 |
+
# Format the prompt for the model
|
16 |
def format_prompt(message, history):
|
17 |
prompt = "<s>"
|
18 |
for user_prompt, bot_response in history:
|
|
|
21 |
prompt += f"[INST] {message} [/INST]"
|
22 |
return prompt
|
23 |
|
24 |
+
# Generate a response from the model
|
25 |
+
def generate(prompt, history=[], temperature=0.1, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0):
|
26 |
+
# Translate the Arabic prompt to English before sending to the model
|
27 |
+
prompt_in_english = translate_to_english(prompt)
|
28 |
+
formatted_prompt = format_prompt(prompt_in_english, history)
|
29 |
+
|
30 |
+
generate_kwargs = {
|
31 |
+
"temperature": temperature,
|
32 |
+
"max_new_tokens": max_new_tokens,
|
33 |
+
"top_p": top_p,
|
34 |
+
"repetition_penalty": repetition_penalty,
|
35 |
+
"do_sample": True,
|
36 |
+
"seed": 42, # Seed for reproducibility, remove or change if randomness is preferred
|
37 |
+
}
|
38 |
+
|
39 |
+
# Generate the response
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
41 |
+
|
42 |
output = ""
|
|
|
43 |
for response in stream:
|
44 |
+
output += response["token"]["text"]
|
45 |
|
46 |
# Translate the English response back to Arabic
|
47 |
+
response_in_arabic = translate_to_arabic(output)
|
48 |
+
return response_in_arabic
|
49 |
+
|
50 |
+
# Define additional inputs for Gradio interface
|
51 |
+
additional_inputs = [
|
52 |
+
gr.Slider(label="Temperature", value=0.9, minimum=0.0, maximum=1.0, step=0.05),
|
53 |
+
gr.Slider(label="Max new tokens", value=256, minimum=0, maximum=1048, step=64),
|
54 |
+
gr.Slider(label="Top-p (nucleus sampling)", value=0.90, minimum=0.0, maximum=1.0, step=0.05),
|
55 |
+
gr.Slider(label="Repetition penalty", value=1.2, minimum=1.0, maximum=2.0, step=0.05)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
]
|
57 |
|
58 |
+
# Set up the Gradio interface
|
59 |
+
iface = gr.Interface(
|
60 |
fn=generate,
|
61 |
+
inputs=[
|
62 |
+
gr.Textbox(lines=5, placeholder='Type your Arabic query here...', label='Arabic Query'),
|
63 |
+
*additional_inputs
|
64 |
+
],
|
65 |
+
outputs='text',
|
66 |
+
title="DorjGPT Arabic-English Translation Chatbot",
|
67 |
+
)
|
68 |
+
|
69 |
+
# Launch the Gradio interface
|
70 |
+
iface.launch()
|