File size: 3,071 Bytes
4e88cbb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 |
import gradio as gr
from transformers import pipeline
from transformers import AutoTokenizer
# Cache for loaded pipelines to avoid reloading
pipeline_cache = {}
# List of available masked language models
def get_model_choices():
return [
"UMCU/CardioMedRoBERTa.nl",
"UMCU/CardioBERTa_base.nl",
"UMCU/CardioBERTa.nl_clinical",
"UMCU/CardioDeBERTa.nl",
"UMCU/CardioDeBERTa.nl_clinical",
#"UMCU/CardioBigBird_base.nl",
"CLTL/MedRoBERTa.nl",
"DTAI-KULeuven/robbert-2023-dutch-base",
"DTAI-KULeuven/robbert-2023-dutch-large",
"joeranbosma/dragon-bert-base-mixed-domain",
"joeranbosma/dragon-bert-base-domain-specific",
"joeranbosma/dragon-roberta-base-mixed-domain",
"joeranbosma/dragon-roberta-large-mixed-domain",
"joeranbosma/dragon-roberta-base-domain-specific",
"joeranbosma/dragon-roberta-large-domain-specific",
"joeranbosma/dragon-longformer-base-mixed-domain",
"joeranbosma/dragon-longformer-large-mixed-domain",
"joeranbosma/dragon-longformer-base-domain-specific",
"joeranbosma/dragon-longformer-large-domain-specific"
]
# Define the prediction function with top-k parameter
def fill_masked(text: str, model_name: str, top_k: int):
"""
Takes text with [MASK] tokens, a model name, and top_k, returns top predictions.
"""
# Load the pipeline if not already cached
if model_name not in pipeline_cache:
pipeline_cache[model_name] = pipeline(
"fill-mask",
model=model_name
)
fill_mask = pipeline_cache[model_name]
# Get top_k predictions
# make sure the mask format is correct
# [MASK] for BERT and DeBERTa
# <mask> for BigBird, LongFormer, RoBERTa and XLM-RoBERTa
#
tokenizer = AutoTokenizer.from_pretrained(model_name)
mask_token = tokenizer.mask_token
text = text.replace("[MASK]", mask_token)
results = fill_mask(text, top_k=top_k)
# Format results for display
formatted = []
for res in results:
formatted.append({
"sequence": res["sequence"],
"score": round(res["score"], 4),
"token": res["token_str"]
})
return formatted
# Build the Gradio interface with a slider for top-k
iface = gr.Interface(
fn=fill_masked,
inputs=[
gr.Textbox(
lines=2,
placeholder="Type text with [MASK] tokens here...",
label="Masked Text"
),
gr.Dropdown(
choices=get_model_choices(),
value="bert-base-uncased",
label="Model"
),
gr.Slider(
minimum=1,
maximum=20,
step=1,
value=5,
label="Top K Predictions"
)
],
outputs=gr.JSON(label="Predictions"),
title="Masked Language Model tester",
description="Enter a sentence with [MASK] tokens, select a model, and choose how many top predictions to return."
)
if __name__ == "__main__":
iface.launch()
|