Add App

emotion/app.py

@@ -0,0 +1,78 @@
+import gradio as gr
+import os
+import torch
+
+from model import create_roberta_model
+from timeit import default_timer as timer
+from typing import Tuple, Dict
+
+# Setup class names
+with open("class_names.txt", "r") as f:
+  class_names = [name.strip() for name in f.readlines()]
+
+### Load example texts ###
+example_texts = []
+with open("example_texts.txt", "r") as file:
+    example_texts = [line.strip() for line in file.readlines()]
+
+### Model and transforms preparation ###
+# Create model and tokenizer
+model, tokenizer = create_roberta_model(output_shape=len(class_names), print_summary=False)
+
+# Load saved weights
+model.load_state_dict(
+    torch.load(f="roberta-base.pth",
+               map_location=torch.device("cpu")) # load to CPU
+)
+
+### Predict function ###
+def predict(text) -> Tuple[Dict, float]:
+  # Start a timer
+  start_time = timer()
+
+  # Set the model to eval
+  model.eval()
+
+  # Set up the inputs
+  inputs = tokenizer(text, padding="max_length", truncation=True, return_tensors='pt')
+
+  # Transform the input image for use with the model
+  X = tokenizer(**inputs).unsqueeze(0) # unsqueeze = add batch dimension on 0th index
+
+  # Put model into eval mode, make prediction
+  model.eval()
+  with torch.inference_mode():
+    # Pass tokenized text through the model and turn the prediction logits into probaiblities
+    pred_probs = torch.softmax(model(X).logits, dim=1)
+
+  # Create a prediction label and prediction probability dictionary
+  pred_labels_and_probs = {class_names[i]: float(pred_probs[0][i]) for i in range(len(class_names))}
+
+  # Calculate pred time
+  end_time = timer()
+  pred_time = round(end_time - start_time, 4)
+
+  # Return pred dict and pred time
+  return pred_labels_and_probs, pred_time
+
+### 4. Gradio app ###
+# Create title, description and article
+title = "A roberta-base Classifier"
+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)"
+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)"
+
+# Create example list
+example_list = [["examples/" + example] for example in os.listdir("examples")]
+
+# Create the Gradio demo
+demo = gr.Interface(fn=predict,
+                    inputs=gr.Textbox(lines=2, placeholder="Type your text here..."),
+                    outputs=[gr.Label(num_top_classes=5, label="Predictions"),
+                             gr.Number(label="Prediction time (s)")],
+                    examples=example_texts,
+                    title=title,
+                    description=description,
+                    article=article)
+
+# Launch the demo
+demo.launch()

emotion/class_names.txt

@@ -0,0 +1,6 @@
+sadness
+joy
+love
+anger
+fear
+surprise

emotion/example_texts.txt

@@ -0,0 +1,3 @@
+I'm feeling blue.
+I hate driving to work!
+I love walking in the park!

emotion/model.py

@@ -0,0 +1,38 @@
+import torch
+from torchinfo import summary
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+def model_input_wrapper(batch_size, sequence_length, tokenizer):
+  dummy_input_ids = torch.randint(0, tokenizer.vocab_size, (batch_size, sequence_length), dtype=torch.long)
+  dummy_attention_mask = torch.ones(batch_size, sequence_length, dtype=torch.long)
+  return {'input_ids': dummy_input_ids, 'attention_mask': dummy_attention_mask}
+
+def create_roberta_model(output_shape:int=10, device=device, print_summary=True):
+  """Creates a HuggingFace roberta-base model.
+
+  Args:
+    device: A torch.device
+    print_summary: A boolean to print the model summary
+
+  Returns:
+    A tuple of the model and tokenizer
+  """
+  tokenizer = AutoTokenizer.from_pretrained('roberta-base')
+  model = AutoModelForSequenceClassification.from_pretrained('roberta-base')
+
+  # Partial Freeze to speed up training
+  for param in model.parameters():
+    param.requires_grad = False
+
+  for param in model.classifier.parameters():
+    param.requires_grad = True
+
+  model.classifier.out_proj = torch.nn.Linear(in_features=768, out_features=output_shape)
+
+  if print_summary:
+    sample_inputs = model_input_wrapper(1, 128, tokenizer)
+    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"]))
+
+  return model.to(device), tokenizer

emotion/requirements.txt

@@ -0,0 +1,4 @@
+torch==2.1.0
+torchvision==0.16.0
+gradio==3.50.2
+transformers==4.35.0

emotion/roberta-base.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5bd0ce37ac5ff9344629a01ecf8395078757659097cee9e3d7f7f7f0ed98f0f
+size 498684833