app.py

import torch
import gradio as gr
from transformers import pipeline, AutoModelForTokenClassification, AutoTokenizer

# Load Model & Tokenizer
model_name = "mmuzamilai/distilbert-drug-ner"
model = AutoModelForTokenClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# NER Pipeline (CPU only)
ner_pipeline = pipeline("ner", model=model, tokenizer=tokenizer, device=-1)

# Label Map
label_map = {
    0: "DOSAGE",
    1: "DRUG_NAME",
    2: "EVENT",
    3: "LOCATION",
    4: "OTHER",
    5: "ROA",
    6: "SYMPTOM",
    7: "TEMPORAL",
}

# Color Map for Pills
color_map = {
    "DOSAGE": "#fbcfe8",     # pink
    "DRUG_NAME": "#a5f3fc",  # sky blue
    "EVENT": "#fde68a",      # yellow
    "LOCATION": "#ddd6fe",   # violet
    "ROA": "#fecaca",        # red
    "SYMPTOM": "#fcd34d",    # amber
    "TEMPORAL": "#c7d2fe",   # indigo
}

def merge_subwords_and_decode(entities):
    merged = []
    current_word = ""
    current_label = None

    for entity in entities:
        word = entity["word"]
        label_id = int(entity["entity"].replace("LABEL_", ""))
        label = label_map.get(label_id, "O")

        if word.startswith("##"):
            current_word += word[2:]
        else:
            if current_word:
                merged.append({"word": current_word, "label": current_label})
            current_word = word
            current_label = label

    if current_word:
        merged.append({"word": current_word, "label": current_label})
    return merged

def format_entities_html(entities):
    """Return HTML with color-coded entity spans."""
    html = ""
    for ent in entities:
        word = ent["word"]
        label = ent["label"]
        if label != "O":
            color = color_map.get(label, "#e5e7eb")
            html += f'<span style="background-color:{color}; padding:2px 6px; margin:2px; border-radius:10px; display:inline-block;">{word} <small style="opacity:0.7">({label})</small></span> '
        else:
            html += f"{word} "
    return html.strip()

def predict_and_format(text):
    raw_entities = ner_pipeline(text)
    cleaned_entities = merge_subwords_and_decode(raw_entities)
    return format_entities_html(cleaned_entities)

# Gradio Interface
demo = gr.Interface(
    fn=predict_and_format,
    inputs=gr.Textbox(lines=4, placeholder="Enter clinical or drug-related text here..."),
    outputs=gr.HTML(label="Named Entity Recognition"),
    title="💊 Drug NER Highlighter",
    description="A custom NER model that highlights drug-related named entities with colorful pills."
)

demo.launch()