Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,92 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
4 |
+
|
5 |
+
# Load the models and tokenizers
|
6 |
+
tokenizer1 = AutoTokenizer.from_pretrained("JasperV13/Fhamator-30000")
|
7 |
+
model1 = AutoModelForCausalLM.from_pretrained("JasperV13/Fhamator-30000")
|
8 |
+
|
9 |
+
tokenizer2 = AutoTokenizer.from_pretrained("JasperV13/Fhamator-SFT")
|
10 |
+
model2 = AutoModelForCausalLM.from_pretrained("JasperV13/Fhamator-SFT")
|
11 |
+
|
12 |
+
def generate_text_fhamator(input_text, max_length, num_return_sequences, no_repeat_ngram_size, top_k, top_p, temperature):
|
13 |
+
input_ids = tokenizer1.encode(input_text, return_tensors='pt')
|
14 |
+
output = model1.generate(
|
15 |
+
input_ids,
|
16 |
+
max_length=max_length,
|
17 |
+
num_return_sequences=num_return_sequences,
|
18 |
+
no_repeat_ngram_size=no_repeat_ngram_size,
|
19 |
+
top_k=top_k,
|
20 |
+
top_p=top_p,
|
21 |
+
temperature=temperature,
|
22 |
+
do_sample=True
|
23 |
+
)
|
24 |
+
generated_texts = [tokenizer1.decode(output[i], skip_special_tokens=True) for i in range(num_return_sequences)]
|
25 |
+
return "\n\n".join(generated_texts)
|
26 |
+
|
27 |
+
def generate_text_sft(input_text, max_length, num_return_sequences, no_repeat_ngram_size, top_k, top_p, temperature):
|
28 |
+
inputs = tokenizer2(input_text, return_tensors="pt")
|
29 |
+
output = model2.generate(
|
30 |
+
inputs['input_ids'],
|
31 |
+
max_length=max_length,
|
32 |
+
num_return_sequences=num_return_sequences,
|
33 |
+
no_repeat_ngram_size=no_repeat_ngram_size,
|
34 |
+
top_k=top_k,
|
35 |
+
top_p=top_p,
|
36 |
+
temperature=temperature,
|
37 |
+
do_sample=True
|
38 |
+
)
|
39 |
+
generated_texts = [tokenizer2.decode(output[i], skip_special_tokens=True) for i in range(num_return_sequences)]
|
40 |
+
return "\n\n".join(generated_texts)
|
41 |
+
|
42 |
+
with gr.Blocks() as demo:
|
43 |
+
with gr.Row():
|
44 |
+
gr.Markdown("## 🤖✨ Fhamator: The Darija-Speaking Model")
|
45 |
+
|
46 |
+
with gr.Tab("📖 Explanation"):
|
47 |
+
gr.Markdown("""
|
48 |
+
# 📚 Explanation of the Work Done
|
49 |
+
Welcome to the **Fhamator** application! This app consists of three main tabs, each with a unique purpose:
|
50 |
+
|
51 |
+
1. **🔍 Explanation**: Provides an overview of what this app does and how it works.
|
52 |
+
2. **🧠 Fhamator Model**: Test the base 'Fhamator-30000' language model and tweak its hyperparameters to see how it performs.
|
53 |
+
3. **🚀 Fhamator-SFT Model**: Experiment with the fine-tuned 'Fhamator-SFT' model, designed for more specific tasks, also with customizable hyperparameters.
|
54 |
+
""")
|
55 |
+
|
56 |
+
with gr.Tab("Test Fhamator-30000 Model"):
|
57 |
+
with gr.Group():
|
58 |
+
input_text = gr.Textbox(label="Input Prompt", value="المغرب هو دولة معروفة ب")
|
59 |
+
max_length = gr.Slider(50, 200, value=100, step=1, label="Max Length")
|
60 |
+
num_return_sequences = gr.Slider(1, 5, value=3, step=1, label="Number of Sequences")
|
61 |
+
no_repeat_ngram_size = gr.Slider(1, 5, value=2, step=1, label="No Repeat N-Gram Size")
|
62 |
+
top_k = gr.Slider(1, 100, value=50, step=1, label="Top K")
|
63 |
+
top_p = gr.Slider(0.0, 1.0, value=0.95, step=0.01, label="Top P")
|
64 |
+
temperature = gr.Slider(0.1, 1.0, value=0.7, step=0.1, label="Temperature")
|
65 |
+
output_text = gr.Textbox(label="Generated Texts", lines=8)
|
66 |
+
generate_btn = gr.Button("Generate")
|
67 |
+
|
68 |
+
generate_btn.click(
|
69 |
+
generate_text_fhamator,
|
70 |
+
inputs=[input_text, max_length, num_return_sequences, no_repeat_ngram_size, top_k, top_p, temperature],
|
71 |
+
outputs=output_text
|
72 |
+
)
|
73 |
+
|
74 |
+
with gr.Tab("Test Fhamator-SFT Model"):
|
75 |
+
with gr.Group():
|
76 |
+
input_text_sft = gr.Textbox(label="Instruction", value="السؤال :س - التاريخ المغربي؟\n\n: الجواب\n")
|
77 |
+
max_length = gr.Slider(50, 200, value=60, step=1, label="Max Length")
|
78 |
+
num_return_sequences = gr.Slider(1, 5, value=5, step=1, label="Number of Sequences")
|
79 |
+
no_repeat_ngram_size = gr.Slider(1, 5, value=2, step=1, label="No Repeat N-Gram Size")
|
80 |
+
top_k_sft = gr.Slider(1, 100, value=50, step=1, label="Top K")
|
81 |
+
top_p_sft = gr.Slider(0.0, 1.0, value=0.95, step=0.01, label="Top P")
|
82 |
+
temperature_sft = gr.Slider(0.1, 1.0, value=0.7, step=0.1, label="Temperature")
|
83 |
+
output_text_sft = gr.Textbox(label="Generated Texts", lines=8)
|
84 |
+
generate_btn_sft = gr.Button("Generate")
|
85 |
+
|
86 |
+
generate_btn_sft.click(
|
87 |
+
generate_text_sft,
|
88 |
+
inputs=[input_text_sft, max_length, num_return_sequences, no_repeat_ngram_size, top_k_sft, top_p_sft, temperature_sft],
|
89 |
+
outputs=output_text_sft
|
90 |
+
)
|
91 |
+
|
92 |
+
demo.launch(debug = True)
|