LukeOLuck commited on
Commit
e418f09
·
1 Parent(s): e7b891a
emotion/app.py ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import os
3
+ import torch
4
+
5
+ from model import create_roberta_model
6
+ from timeit import default_timer as timer
7
+ from typing import Tuple, Dict
8
+
9
+ # Setup class names
10
+ with open("class_names.txt", "r") as f:
11
+ class_names = [name.strip() for name in f.readlines()]
12
+
13
+ ### Load example texts ###
14
+ example_texts = []
15
+ with open("example_texts.txt", "r") as file:
16
+ example_texts = [line.strip() for line in file.readlines()]
17
+
18
+ ### Model and transforms preparation ###
19
+ # Create model and tokenizer
20
+ model, tokenizer = create_roberta_model(output_shape=len(class_names), print_summary=False)
21
+
22
+ # Load saved weights
23
+ model.load_state_dict(
24
+ torch.load(f="roberta-base.pth",
25
+ map_location=torch.device("cpu")) # load to CPU
26
+ )
27
+
28
+ ### Predict function ###
29
+ def predict(text) -> Tuple[Dict, float]:
30
+ # Start a timer
31
+ start_time = timer()
32
+
33
+ # Set the model to eval
34
+ model.eval()
35
+
36
+ # Set up the inputs
37
+ inputs = tokenizer(text, padding="max_length", truncation=True, return_tensors='pt')
38
+
39
+ # Transform the input image for use with the model
40
+ X = tokenizer(**inputs).unsqueeze(0) # unsqueeze = add batch dimension on 0th index
41
+
42
+ # Put model into eval mode, make prediction
43
+ model.eval()
44
+ with torch.inference_mode():
45
+ # Pass tokenized text through the model and turn the prediction logits into probaiblities
46
+ pred_probs = torch.softmax(model(X).logits, dim=1)
47
+
48
+ # Create a prediction label and prediction probability dictionary
49
+ pred_labels_and_probs = {class_names[i]: float(pred_probs[0][i]) for i in range(len(class_names))}
50
+
51
+ # Calculate pred time
52
+ end_time = timer()
53
+ pred_time = round(end_time - start_time, 4)
54
+
55
+ # Return pred dict and pred time
56
+ return pred_labels_and_probs, pred_time
57
+
58
+ ### 4. Gradio app ###
59
+ # Create title, description and article
60
+ title = "A roberta-base Classifier"
61
+ description = "[A roberta-base BERT based model](https://huggingface.co/roberta-base) text model to classify text on the [HuggingFace 🤗 dair-ai/emotion dataset](https://huggingface.co/datasets/dair-ai/emotion). [Source Code Found Here](https://colab.research.google.com/drive/1P7rfiDF1jfNHKmkB7WjHPi8PQBLQ4Ege?usp=sharing)"
62
+ article = "Built with [Gradio](https://github.com/gradio-app/gradio) and [PyTorch](https://pytorch.org/). [Source Code Found Here](https://colab.research.google.com/drive/1P7rfiDF1jfNHKmkB7WjHPi8PQBLQ4Ege?usp=sharing)"
63
+
64
+ # Create example list
65
+ example_list = [["examples/" + example] for example in os.listdir("examples")]
66
+
67
+ # Create the Gradio demo
68
+ demo = gr.Interface(fn=predict,
69
+ inputs=gr.Textbox(lines=2, placeholder="Type your text here..."),
70
+ outputs=[gr.Label(num_top_classes=5, label="Predictions"),
71
+ gr.Number(label="Prediction time (s)")],
72
+ examples=example_texts,
73
+ title=title,
74
+ description=description,
75
+ article=article)
76
+
77
+ # Launch the demo
78
+ demo.launch()
emotion/class_names.txt ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ sadness
2
+ joy
3
+ love
4
+ anger
5
+ fear
6
+ surprise
emotion/example_texts.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ I'm feeling blue.
2
+ I hate driving to work!
3
+ I love walking in the park!
emotion/model.py ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ from torchinfo import summary
3
+ from transformers import AutoTokenizer, AutoModelForSequenceClassification
4
+
5
+ device = "cuda" if torch.cuda.is_available() else "cpu"
6
+
7
+ def model_input_wrapper(batch_size, sequence_length, tokenizer):
8
+ dummy_input_ids = torch.randint(0, tokenizer.vocab_size, (batch_size, sequence_length), dtype=torch.long)
9
+ dummy_attention_mask = torch.ones(batch_size, sequence_length, dtype=torch.long)
10
+ return {'input_ids': dummy_input_ids, 'attention_mask': dummy_attention_mask}
11
+
12
+ def create_roberta_model(output_shape:int=10, device=device, print_summary=True):
13
+ """Creates a HuggingFace roberta-base model.
14
+
15
+ Args:
16
+ device: A torch.device
17
+ print_summary: A boolean to print the model summary
18
+
19
+ Returns:
20
+ A tuple of the model and tokenizer
21
+ """
22
+ tokenizer = AutoTokenizer.from_pretrained('roberta-base')
23
+ model = AutoModelForSequenceClassification.from_pretrained('roberta-base')
24
+
25
+ # Partial Freeze to speed up training
26
+ for param in model.parameters():
27
+ param.requires_grad = False
28
+
29
+ for param in model.classifier.parameters():
30
+ param.requires_grad = True
31
+
32
+ model.classifier.out_proj = torch.nn.Linear(in_features=768, out_features=output_shape)
33
+
34
+ if print_summary:
35
+ sample_inputs = model_input_wrapper(1, 128, tokenizer)
36
+ print(summary(model, input_data=sample_inputs, verbose=0, col_names=["input_size", "output_size", "num_params", "trainable"], col_width=20, row_settings=["var_names"]))
37
+
38
+ return model.to(device), tokenizer
emotion/requirements.txt ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ torch==2.1.0
2
+ torchvision==0.16.0
3
+ gradio==3.50.2
4
+ transformers==4.35.0
emotion/roberta-base.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e5bd0ce37ac5ff9344629a01ecf8395078757659097cee9e3d7f7f7f0ed98f0f
3
+ size 498684833