import gradio as gr
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
from huggingface_hub import login

# Dictionary of AI artifacts with explanations
artifact_definitions = {
    "Repetition": "The same word or phrase is repeated unnaturally within the text. AI struggles to generate new, diverse text.",
    "Incoherent Phrases": "Sentences or phrases that don’t make sense when put together. Lack of contextual understanding.",
    "Ambiguous Pronouns": "Unclear references where pronouns like 'it' or 'they' don’t correspond clearly to any noun. Weakness in coreference resolution.",
    "Overuse of Filler Words": "Frequent use of words like 'however,' 'therefore,' or 'thus,' which don’t contribute to the meaning of the text. Model generates overly formal language.",
    "Generic Responses": "Vague answers or phrases that don’t provide meaningful information, such as 'it depends' or 'this is important.' Avoids specificity due to lack of information.",
    "Contradictions": "Statements in the generated text that contradict each other. Model struggles with logical consistency.",
    "Unnatural Syntax": "Phrases that sound robotic or awkward compared to natural human language. Incomplete grasp of human grammar nuances.",
    "Template-like Responses": "Responses that follow a rigid, repetitive format rather than natural flow. Over-reliance on training data patterns.",
    "Unnecessary Formality": "Overly formal or stiff language that feels unnatural in casual contexts. AI doesn’t adapt to tone appropriately.",
    "Incorrect Facts": "Inclusion of wrong or outdated information that was not verified. Lack of fact-checking abilities.",
    "Hallucinations": "Content or information generated by AI that has no basis in the input or training data. Model invents false information.",
    "Clichéd Phrases": "Overuse of common phrases or idioms that seem out of place in the text. Model mirrors over-represented text data.",
    "Overuse of Adverbs": "Excessive use of words like 'really,' 'very,' or 'extremely,' which don’t add to the meaning of the sentence. Attempts to mimic human emphasis poorly.",
    "Inconsistent Terminology": "Shifting between different terms for the same concept within a short text. Lack of term consistency management.",
    "Overgeneralization": "Statements that are too broad or generic, such as 'all people' or 'always.' AI fails to account for nuances or specifics.",
    "Lack of Subject-Specific Knowledge": "Responses that avoid deeper understanding of specialized topics. Insufficient training in specific domains.",
    "Awkward Phrasing": "Sentences that sound unnatural due to word order or selection. Lack of nuanced syntactic understanding.",
    "Redundancy": "Repetition of information or points already made earlier in the text. Fails to maintain conciseness.",
    "Unintended Humor": "Sentences that are accidentally funny due to odd phrasing or unexpected associations. AI misunderstanding of language subtleties."
}

# Load the model from Hugging Face
# Replace with the actual username and repository name of your model
model_name_or_path = "facebook/bart-large-mnli"  # Example: Replace with your model repo name
model = AutoModelForSequenceClassification.from_pretrained(model_name_or_path)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)

# Create a pipeline for text classification
artifact_classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)

# Function to detect AI artifacts and provide explanations
def detect_artifacts(text):
    results = artifact_classifier(text)
    response = []
    for result in results:
        label = result['label']
        score = result['score']
        explanation = artifact_definitions.get(label, "No explanation available.")
        response.append(f"**Artifact Detected**: {label}\n**Confidence**: {score:.2f}\n**Explanation**: {explanation}")
    return "\n\n".join(response)

# Gradio interface setup
interface = gr.Interface(
    fn=detect_artifacts,
    inputs="text",
    outputs="text",
    title="AI Artifact Detection",
    description=(
        "Paste AI-generated text to detect common AI artifacts, such as repetition, "
        "contradictions, or unnatural syntax. The app will also explain why these artifacts occur."
    ),
    examples=[
        ["However, the issue was repeated multiple times, making it confusing and unclear."],
        ["The process was very redundant and robotic."],
        ["At the end of the day, this contradiction was unavoidable."]
    ]
)

# Launch the Gradio app
if __name__ == "__main__":
    interface.launch()