Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,13 @@
|
|
1 |
import gradio as gr
|
2 |
from PIL import Image
|
3 |
-
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer,
|
4 |
|
5 |
import soundfile as sf
|
6 |
import torch
|
7 |
import numpy as np
|
|
|
8 |
|
|
|
9 |
class_names = {
|
10 |
0: "al qarawiyyin",
|
11 |
1: "bab mansour el aleuj",
|
@@ -16,94 +18,79 @@ class_names = {
|
|
16 |
6: "madrasa ben youssef",
|
17 |
7: "majorel gardens",
|
18 |
8: "menara"
|
19 |
-
|
20 |
|
21 |
model_name_or_path = "microsoft/DialoGPT-large"
|
22 |
-
|
23 |
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, padding_side="left", use_fast=False)
|
24 |
tokenizer.pad_token = tokenizer.eos_token
|
|
|
25 |
|
26 |
-
|
27 |
-
|
28 |
-
torch_dtype=torch.float32,
|
29 |
-
device_map="auto",
|
30 |
-
trust_remote_code=True,
|
31 |
-
)
|
32 |
-
|
33 |
-
# Initialize the Wav2Vec2 model and processor
|
34 |
-
wav2vec2_processor = WhisperProcessor.from_pretrained("openai/whisper-large")
|
35 |
-
wav2vec2_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large")
|
36 |
-
wav2vec2_model.config.forced_decoder_ids = None
|
37 |
|
38 |
vit_model = ViTForImageClassification.from_pretrained('ohidaoui/monuments-morocco-v1')
|
39 |
vit_feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-base-patch16-224')
|
40 |
|
41 |
-
|
42 |
# Function to handle text input
|
43 |
def handle_text(text):
|
44 |
-
|
45 |
-
|
46 |
-
chat_history_ids = model.generate(bot_input_ids, max_length=1000, pad_token_id=tokenizer.eos_token_id)
|
47 |
-
chat_output = tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True)
|
48 |
-
return chat_output
|
49 |
|
50 |
# Function to handle image input
|
51 |
def get_class_name(class_idx):
|
52 |
return class_names[class_idx]
|
53 |
|
54 |
-
|
55 |
def handle_image(img):
|
56 |
-
# Convert PIL image to numpy array
|
57 |
img = np.array(img)
|
58 |
-
|
59 |
-
# Apply transformations and prepare image for the model
|
60 |
inputs = vit_feature_extractor(images=img, return_tensors="pt")
|
61 |
-
|
62 |
-
# Pass through the Vision Transformer model
|
63 |
outputs = vit_model(**inputs)
|
64 |
-
|
65 |
-
# Get the predicted class
|
66 |
predicted_class_idx = torch.argmax(outputs.logits, dim=1).item()
|
67 |
-
|
68 |
-
|
69 |
predicted_class_name = get_class_name(predicted_class_idx)
|
|
|
|
|
70 |
|
71 |
-
return predicted_class_name
|
72 |
-
|
73 |
# Function to handle audio input
|
74 |
def handle_audio(audio):
|
75 |
-
# gradio Audio returns a tuple (sample_rate, audio_np_array)
|
76 |
-
# we only need the audio data, hence accessing the second element
|
77 |
audio = audio[1]
|
78 |
-
input_values = wav2vec2_processor(audio, sampling_rate=
|
79 |
-
# Convert to the expected tensor type
|
80 |
input_values = input_values.to(torch.float32)
|
81 |
logits = wav2vec2_model(input_values).logits
|
82 |
predicted_ids = torch.argmax(logits, dim=-1)
|
83 |
transcriptions = wav2vec2_processor.decode(predicted_ids[0])
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
|
89 |
-
|
|
|
90 |
text_output = handle_text(text) if text is not None else ''
|
91 |
img_output = handle_image(img) if img is not None else ''
|
92 |
audio_output = handle_audio(audio) if audio is not None else ''
|
93 |
-
|
94 |
outputs = [o for o in [text_output, img_output, audio_output] if o]
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
)
|
108 |
-
|
109 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
from PIL import Image
|
3 |
+
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer, Wav2Vec2Processor, Wav2Vec2ForCTC, ViTFeatureExtractor, ViTForImageClassification
|
4 |
|
5 |
import soundfile as sf
|
6 |
import torch
|
7 |
import numpy as np
|
8 |
+
import time
|
9 |
|
10 |
+
# Initialize the transformers and the models
|
11 |
class_names = {
|
12 |
0: "al qarawiyyin",
|
13 |
1: "bab mansour el aleuj",
|
|
|
18 |
6: "madrasa ben youssef",
|
19 |
7: "majorel gardens",
|
20 |
8: "menara"
|
21 |
+
}
|
22 |
|
23 |
model_name_or_path = "microsoft/DialoGPT-large"
|
|
|
24 |
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, padding_side="left", use_fast=False)
|
25 |
tokenizer.pad_token = tokenizer.eos_token
|
26 |
+
model = AutoModelForCausalLM.from_pretrained(model_name_or_path, torch_dtype=torch.float32, trust_remote_code=True)
|
27 |
|
28 |
+
wav2vec2_processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
|
29 |
+
wav2vec2_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
vit_model = ViTForImageClassification.from_pretrained('ohidaoui/monuments-morocco-v1')
|
32 |
vit_feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-base-patch16-224')
|
33 |
|
|
|
34 |
# Function to handle text input
|
35 |
def handle_text(text):
|
36 |
+
chat_output = chat({"question": text})
|
37 |
+
return chat_output["answer"]
|
|
|
|
|
|
|
38 |
|
39 |
# Function to handle image input
|
40 |
def get_class_name(class_idx):
|
41 |
return class_names[class_idx]
|
42 |
|
|
|
43 |
def handle_image(img):
|
|
|
44 |
img = np.array(img)
|
|
|
|
|
45 |
inputs = vit_feature_extractor(images=img, return_tensors="pt")
|
|
|
|
|
46 |
outputs = vit_model(**inputs)
|
|
|
|
|
47 |
predicted_class_idx = torch.argmax(outputs.logits, dim=1).item()
|
|
|
|
|
48 |
predicted_class_name = get_class_name(predicted_class_idx)
|
49 |
+
chat_output = chat({"question": "what is " + predicted_class_name})
|
50 |
+
return chat_output["answer"]
|
51 |
|
|
|
|
|
52 |
# Function to handle audio input
|
53 |
def handle_audio(audio):
|
|
|
|
|
54 |
audio = audio[1]
|
55 |
+
input_values = wav2vec2_processor(audio, sampling_rate=16_000, return_tensors="pt").input_values
|
|
|
56 |
input_values = input_values.to(torch.float32)
|
57 |
logits = wav2vec2_model(input_values).logits
|
58 |
predicted_ids = torch.argmax(logits, dim=-1)
|
59 |
transcriptions = wav2vec2_processor.decode(predicted_ids[0])
|
60 |
+
chat_output = chat({"question": transcriptions})
|
61 |
+
return chat_output["answer"]
|
|
|
|
|
62 |
|
63 |
+
# Main function to handle the inputs
|
64 |
+
def chatbot(history, text=None, img=None, audio=None):
|
65 |
text_output = handle_text(text) if text is not None else ''
|
66 |
img_output = handle_image(img) if img is not None else ''
|
67 |
audio_output = handle_audio(audio) if audio is not None else ''
|
|
|
68 |
outputs = [o for o in [text_output, img_output, audio_output] if o]
|
69 |
+
output = "\n".join(outputs)
|
70 |
+
|
71 |
+
history[-1][1] = output
|
72 |
+
for character in output:
|
73 |
+
history[-1][1] += character
|
74 |
+
time.sleep(0.05)
|
75 |
+
yield history
|
76 |
+
|
77 |
+
with gr.Blocks() as demo:
|
78 |
+
chat_interface = gr.Chatbot([], elem_id="chatbot", height=750)
|
79 |
+
|
80 |
+
with gr.Row():
|
81 |
+
with gr.Column(scale=0.85):
|
82 |
+
text_input = gr.Textbox(
|
83 |
+
show_label=False,
|
84 |
+
placeholder="Input Text here...",
|
85 |
+
container=False
|
86 |
+
)
|
87 |
+
with gr.Column(scale=0.15, min_width=0):
|
88 |
+
img_input = gr.Image()
|
89 |
+
audio_input = gr.Audio(source="microphone", label="Audio Input")
|
90 |
+
|
91 |
+
text_msg = text_input.submit(chatbot, [chat_interface, text_input], [chat_interface, text_input], queue=False)
|
92 |
+
img_msg = img_input.upload(chatbot, [chat_interface, img_input], [chat_interface, img_input], queue=False)
|
93 |
+
audio_msg = audio_input.upload(chatbot, [chat_interface, audio_input], [chat_interface, audio_input], queue=False)
|
94 |
+
|
95 |
+
demo.queue()
|
96 |
+
demo.launch(share=True)
|