# HuggingFace Spaces file to run a Gradio Interface for the ALBERT v2 Steam Review Constructiveness Classifier by Samuel Ruairí Bullard # Package Imports import gradio as gr from transformers import pipeline import torch # Checks if CUDA is available on the machine print("CUDA Available: ", torch.cuda.is_available()) # if not os.path.isfile("./README.md"): # !git clone https://huggingface.co/spaces/abullard1/albert-v2-steam-review-constructiveness-classifier # Sets the torch dtype to 16-bit half-precision floating-point format if CUDA is available, otherwise sets it to 32-bit single-precision floating-point format. (Available for GPUs with Tensor Cores like NVIDIA's Volta, Turing, Ampere Architectures have for example) device = 0 if torch.cuda.is_available() else -1 torch_d_type = torch.float16 if torch.cuda.is_available() else torch.float32 print(f"Device: {device}") # Defines the name of the base model, the classifier was fine-tuned from base_model_name = "albert-base-v2" # Defines the name of the fine-tuned model used for the steam-review constructiveness classification finetuned_model_name = "abullard1/albert-v2-steam-review-constructiveness-classifier" # PyTorch classifier pipeline classifier = pipeline( task="text-classification", # Defines the task model=finetuned_model_name, # Defines the fine-tuned model to use tokenizer=base_model_name, # Defines the tokenizer to use (same as the base model) device=device, # Defines the device the classification will be run on top_k=None, # Returns all scores for all labels, not just the one with the highest score truncation=True, # Truncates the input text if it exceeds the maximum length max_length=512, # Defines the maximum length of the input text (512 for BERT. Explicitly set here) torch_dtype=torch_d_type # Sets the torch dtype to 16-bit half-precision floating-point format if CUDA is available, otherwise sets it to 32-bit single-precision floating-point format ) # Extracts the labels and scores from the prediction result def classify_steam_review(input_text): result = classifier(input_text) label_1, label_2 = result[0][0]["label"], result[0][1]["label"] score_1, score_2 = round(result[0][0]["score"], 6), round(result[0][1]["score"], 6) return {"label_1": label_1, "score_1": score_1, "label_2": label_2, "score_2": score_2} # Provides a textual representation of the classification result def get_steam_review_classification_result_text(label_1, score_1): if label_1 == "LABEL_1": return f"Constructive with a score of {score_1}. 👍🏻" else: return f"Not Constructive with a score of {score_1}. 👎🏻" # Examples Steam Reviews to display in the Gradio Interface using the "examples" parameter examples = [ ["Review: I think this is a great game but it still has some room for improvement., Playtime: 12, Voted Up: True, Upvotes: 1, Votes Funny: 0"], ["Review: Trash game. Deleted., Playtime: 1, Voted Up: False, Upvotes: 0, Votes Funny: 0"], ["Review: This game is amazing., Playtime: 100, Voted Up: True, Upvotes: 1, Votes Funny: 0"], ["Review: Great game, but the community is toxic., Playtime: 50, Voted Up: True, Upvotes: 1, Votes Funny: 0"] ] # Information Description about the Steam Review Format # info_description = ( # """ # Format your input as follows for the best results: ***Review**: {review_text}, **Playtime**: {author_playtime_at_review}, **Voted Up**: {voted_up}, **Upvotes**: {upvotes}, **Votes Funny**: {votes_funny}.* # """ # ) # Labeling Criteria Information Markdown labeling_criteria = ( """