Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -11,9 +11,8 @@ from transformers import (
|
|
11 |
from scipy.io.wavfile import write
|
12 |
import tempfile
|
13 |
from dotenv import load_dotenv
|
14 |
-
import spaces
|
15 |
|
16 |
-
# Load environment variables (e.g., Hugging Face token)
|
17 |
load_dotenv()
|
18 |
hf_token = os.getenv("HF_TOKEN")
|
19 |
|
@@ -21,7 +20,7 @@ hf_token = os.getenv("HF_TOKEN")
|
|
21 |
# ---------------------------------------------------------------------
|
22 |
# Load Llama 3 Pipeline with Zero GPU (Encapsulated)
|
23 |
# ---------------------------------------------------------------------
|
24 |
-
@spaces.GPU(duration=300)
|
25 |
def generate_script(user_prompt: str, model_id: str, token: str):
|
26 |
try:
|
27 |
tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
|
@@ -35,9 +34,10 @@ def generate_script(user_prompt: str, model_id: str, token: str):
|
|
35 |
llama_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
36 |
|
37 |
system_prompt = (
|
38 |
-
|
39 |
-
|
40 |
-
|
|
|
41 |
combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nRefined script:"
|
42 |
result = llama_pipeline(combined_prompt, max_new_tokens=200, do_sample=True, temperature=0.9)
|
43 |
return result[0]["generated_text"].split("Refined script:")[-1].strip()
|
@@ -85,23 +85,63 @@ def interface_generate_audio(script, audio_length):
|
|
85 |
# Interface
|
86 |
# ---------------------------------------------------------------------
|
87 |
with gr.Blocks() as demo:
|
88 |
-
|
89 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
with gr.Row():
|
91 |
user_prompt = gr.Textbox(
|
92 |
-
label="Enter
|
93 |
-
placeholder="E.g., A 15-second
|
|
|
|
|
94 |
)
|
95 |
llama_model_id = gr.Textbox(
|
96 |
-
label="Llama 3 Model ID",
|
|
|
|
|
97 |
)
|
98 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
|
|
|
|
|
|
104 |
|
|
|
105 |
generate_script_button.click(
|
106 |
fn=interface_generate_script,
|
107 |
inputs=[user_prompt, llama_model_id],
|
|
|
11 |
from scipy.io.wavfile import write
|
12 |
import tempfile
|
13 |
from dotenv import load_dotenv
|
14 |
+
import spaces
|
15 |
|
|
|
16 |
load_dotenv()
|
17 |
hf_token = os.getenv("HF_TOKEN")
|
18 |
|
|
|
20 |
# ---------------------------------------------------------------------
|
21 |
# Load Llama 3 Pipeline with Zero GPU (Encapsulated)
|
22 |
# ---------------------------------------------------------------------
|
23 |
+
@spaces.GPU(duration=300)
|
24 |
def generate_script(user_prompt: str, model_id: str, token: str):
|
25 |
try:
|
26 |
tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
|
|
|
34 |
llama_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
35 |
|
36 |
system_prompt = (
|
37 |
+
"You are an expert radio imaging producer specializing in sound design and music. "
|
38 |
+
"Take the user's concept and craft a concise, creative promo script with a strong focus on auditory elements and musical appeal."
|
39 |
+
)
|
40 |
+
|
41 |
combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nRefined script:"
|
42 |
result = llama_pipeline(combined_prompt, max_new_tokens=200, do_sample=True, temperature=0.9)
|
43 |
return result[0]["generated_text"].split("Refined script:")[-1].strip()
|
|
|
85 |
# Interface
|
86 |
# ---------------------------------------------------------------------
|
87 |
with gr.Blocks() as demo:
|
88 |
+
# Header
|
89 |
+
gr.Markdown("""
|
90 |
+
# ποΈ AI-Powered Radio Imaging Studio π
|
91 |
+
### Create stunning **radio promos** with **Llama 3** and **MusicGen**
|
92 |
+
π₯ **Zero GPU** integration for efficiency and ease!
|
93 |
+
""")
|
94 |
+
|
95 |
+
# Script Generation Section
|
96 |
+
gr.Markdown("## βοΈ Step 1: Generate Your Promo Script")
|
97 |
with gr.Row():
|
98 |
user_prompt = gr.Textbox(
|
99 |
+
label="π€ Enter Promo Idea",
|
100 |
+
placeholder="E.g., A 15-second energetic jingle for a morning talk show.",
|
101 |
+
lines=2,
|
102 |
+
info="Describe your promo idea clearly to generate a creative script."
|
103 |
)
|
104 |
llama_model_id = gr.Textbox(
|
105 |
+
label="ποΈ Llama 3 Model ID",
|
106 |
+
value="meta-llama/Meta-Llama-3-8B-Instruct",
|
107 |
+
info="Enter the Hugging Face model ID for Llama 3."
|
108 |
)
|
109 |
+
generate_script_button = gr.Button("Generate Script β¨")
|
110 |
+
script_output = gr.Textbox(
|
111 |
+
label="π Generated Promo Script",
|
112 |
+
lines=4,
|
113 |
+
interactive=False,
|
114 |
+
info="Your generated promo script will appear here."
|
115 |
+
)
|
116 |
+
|
117 |
+
# Audio Generation Section
|
118 |
+
gr.Markdown("## π§ Step 2: Generate Audio from Your Script")
|
119 |
+
with gr.Row():
|
120 |
+
audio_length = gr.Slider(
|
121 |
+
label="π΅ Audio Length (tokens)",
|
122 |
+
minimum=128,
|
123 |
+
maximum=1024,
|
124 |
+
step=64,
|
125 |
+
value=512,
|
126 |
+
info="Select the desired audio token length."
|
127 |
+
)
|
128 |
+
generate_audio_button = gr.Button("Generate Audio πΆ")
|
129 |
+
audio_output = gr.Audio(
|
130 |
+
label="πΆ Generated Audio File",
|
131 |
+
type="filepath",
|
132 |
+
interactive=False,
|
133 |
+
info="Listen to the audio file generated from your script."
|
134 |
+
)
|
135 |
|
136 |
+
# Footer
|
137 |
+
gr.Markdown("""
|
138 |
+
<br><hr>
|
139 |
+
<p style="text-align: center; font-size: 0.9em;">
|
140 |
+
Created with β€οΈ by <a href="https://bilsimaging.com" target="_blank">bilsimaging.com</a>
|
141 |
+
</p>
|
142 |
+
""", elem_id="footer")
|
143 |
|
144 |
+
# Button Actions
|
145 |
generate_script_button.click(
|
146 |
fn=interface_generate_script,
|
147 |
inputs=[user_prompt, llama_model_id],
|