Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,109 +1,86 @@
|
|
1 |
import torch
|
|
|
|
|
|
|
|
|
|
|
2 |
|
|
|
|
|
3 |
|
|
|
4 |
device = "cpu"
|
5 |
-
model_id ="ALLaM-AI/ALLaM-7B-Instruct-preview"
|
6 |
-
|
7 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
8 |
-
|
9 |
|
|
|
10 |
model = AutoModelForCausalLM.from_pretrained(
|
11 |
model_id,
|
12 |
torch_dtype="auto",
|
13 |
trust_remote_code=True,
|
14 |
)
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
return_full_text=False,
|
28 |
-
max_new_tokens=500,
|
29 |
-
do_sample=False
|
30 |
)
|
31 |
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
pipeline = KPipeline(lang_code='b', model=False)
|
37 |
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
|
|
40 |
def Generate_audio(text, voice='bm_lewis', speed=1):
|
41 |
-
|
42 |
pipeline = KPipeline(lang_code='b')
|
43 |
-
|
44 |
-
|
45 |
generator = pipeline(text, voice=voice, speed=speed, split_pattern=r'\n+')
|
46 |
-
|
47 |
-
|
48 |
full_audio = []
|
49 |
for _, _, audio in generator:
|
50 |
full_audio.extend(audio)
|
51 |
-
|
52 |
-
|
53 |
full_audio = np.array(full_audio)
|
54 |
-
|
55 |
-
|
56 |
return full_audio, 24000
|
57 |
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
captionImage = transformers_pipeline("image-to-text",
|
63 |
-
model="Salesforce/blip-image-captioning-large")
|
64 |
-
|
65 |
-
def Image_Caption(image):
|
66 |
-
caption = captionImage(image)
|
67 |
-
caption = caption[0]['generated_text']
|
68 |
-
return caption
|
69 |
-
|
70 |
-
def Generate_story(textAbout):
|
71 |
-
storyAbout = {"role": "user", "content": f'write a long story about {textAbout} that takes 3 min to read'},
|
72 |
-
|
73 |
-
story = generator(storyAbout)
|
74 |
-
story = story[0]['generated_text']
|
75 |
-
story = story.replace('\n', ' ').replace('arafed', ' ')
|
76 |
-
return story
|
77 |
-
|
78 |
def Mustalhim(image):
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
|
|
|
84 |
def gradio_interface(image):
|
85 |
-
|
86 |
audio_waveform, sampling_rate = Mustalhim(image)
|
87 |
-
|
88 |
-
|
89 |
audio_file = "output_audio.wav"
|
90 |
sf.write(audio_file, audio_waveform, sampling_rate)
|
91 |
-
|
92 |
-
|
93 |
return audio_file
|
94 |
|
|
|
95 |
example_image = "Example.PNG"
|
96 |
|
97 |
-
|
98 |
app = gr.Interface(
|
99 |
-
fn=gradio_interface,
|
100 |
-
inputs=gr.Image(type="pil"),
|
101 |
-
outputs=gr.Audio(type="filepath"),
|
102 |
title="Image to Audio Story",
|
103 |
description="Upload an image, and the app will generate a story and convert it to audio.",
|
104 |
examples=[[example_image]]
|
105 |
)
|
106 |
|
107 |
# Launch the app
|
108 |
-
app.launch()
|
109 |
-
|
|
|
1 |
import torch
|
2 |
+
from transformers import AutoModelForCausalLM, LlamaTokenizer, pipeline as transformers_pipeline
|
3 |
+
from kokoro import KPipeline
|
4 |
+
import soundfile as sf
|
5 |
+
import numpy as np
|
6 |
+
import gradio as gr
|
7 |
|
8 |
+
# Initialize the image-to-text pipeline
|
9 |
+
captionImage = transformers_pipeline("image-to-text", model="Salesforce/blip-image-captioning-large")
|
10 |
|
11 |
+
# Initialize the text-generation pipeline
|
12 |
device = "cpu"
|
13 |
+
model_id = "ALLaM-AI/ALLaM-7B-Instruct-preview"
|
|
|
|
|
|
|
14 |
|
15 |
+
# Load the model
|
16 |
model = AutoModelForCausalLM.from_pretrained(
|
17 |
model_id,
|
18 |
torch_dtype="auto",
|
19 |
trust_remote_code=True,
|
20 |
)
|
21 |
|
22 |
+
# Use LlamaTokenizer for compatibility
|
23 |
+
tokenizer = LlamaTokenizer.from_pretrained(model_id)
|
24 |
+
|
25 |
+
# Initialize the text-generation pipeline
|
26 |
+
generator = transformers_pipeline(
|
27 |
+
"text-generation",
|
28 |
+
model=model,
|
29 |
+
tokenizer=tokenizer,
|
30 |
+
return_full_text=False,
|
31 |
+
max_new_tokens=500,
|
32 |
+
do_sample=False,
|
|
|
|
|
|
|
33 |
)
|
34 |
|
35 |
+
# Function to generate caption
|
36 |
+
def Image_Caption(image):
|
37 |
+
caption = captionImage(image)
|
38 |
+
return caption[0]['generated_text']
|
|
|
39 |
|
40 |
+
# Function to generate a story
|
41 |
+
def Generate_story(textAbout):
|
42 |
+
storyAbout = {"role": "user", "content": f'write a long story about {textAbout} that takes 3 min to read'}
|
43 |
+
story = generator(storyAbout)
|
44 |
+
story = story[0]['generated_text']
|
45 |
+
story = story.replace('\n', ' ').replace('arafed', ' ')
|
46 |
+
return story
|
47 |
|
48 |
+
# Function to generate audio
|
49 |
def Generate_audio(text, voice='bm_lewis', speed=1):
|
|
|
50 |
pipeline = KPipeline(lang_code='b')
|
|
|
|
|
51 |
generator = pipeline(text, voice=voice, speed=speed, split_pattern=r'\n+')
|
|
|
|
|
52 |
full_audio = []
|
53 |
for _, _, audio in generator:
|
54 |
full_audio.extend(audio)
|
|
|
|
|
55 |
full_audio = np.array(full_audio)
|
|
|
|
|
56 |
return full_audio, 24000
|
57 |
|
58 |
+
# Main function to process the image and generate audio
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
def Mustalhim(image):
|
60 |
+
caption = Image_Caption(image)
|
61 |
+
story = Generate_story(caption)
|
62 |
+
audio = Generate_audio(story)
|
63 |
+
return audio
|
64 |
|
65 |
+
# Gradio interface
|
66 |
def gradio_interface(image):
|
|
|
67 |
audio_waveform, sampling_rate = Mustalhim(image)
|
|
|
|
|
68 |
audio_file = "output_audio.wav"
|
69 |
sf.write(audio_file, audio_waveform, sampling_rate)
|
|
|
|
|
70 |
return audio_file
|
71 |
|
72 |
+
# Path to the example image
|
73 |
example_image = "Example.PNG"
|
74 |
|
75 |
+
# Create the Gradio app
|
76 |
app = gr.Interface(
|
77 |
+
fn=gradio_interface,
|
78 |
+
inputs=gr.Image(type="pil"),
|
79 |
+
outputs=gr.Audio(type="filepath"),
|
80 |
title="Image to Audio Story",
|
81 |
description="Upload an image, and the app will generate a story and convert it to audio.",
|
82 |
examples=[[example_image]]
|
83 |
)
|
84 |
|
85 |
# Launch the app
|
86 |
+
app.launch()
|
|