maliozer's picture
Update app.py (#1)
e08ef6b
import torch
import gradio as gr
from transformers import BioGptTokenizer, BioGptForCausalLM
model_names = [
"BioGPT",
"BioGPT-Large",
"BioGPT-QA-PubMedQA-BioGPT",
"BioGPT-QA-PubMEDQA-BioGPT-Large",
"BioGPT-RE-BC5CDR",
"BioGPT-RE-DDI",
"BioGPT-RE-DTI",
"BioGPT-DC-HoC"
]
def load_model(model_name="BioGPT"):
model_name_map = {
"BioGPT":"microsoft/biogpt",
"BioGPT-QA-PubMedQA-BioGPT":"microsoft/BioGPT-Large-PubMedQA"
}
tokenizer = BioGptTokenizer.from_pretrained(model_name_map[model_name])
model = BioGptForCausalLM.from_pretrained(model_name_map[model_name])
return tokenizer, model
def get_beam_output(sentence, selected_model, min_len=100,max_len=512, n_beams=1):
tokenizer, model = load_model(selected_model)
inputs = tokenizer(sentence, return_tensors="pt")
with torch.no_grad():
beam_output = model.generate(**inputs,
min_length=min_len,
max_length=max_len,
num_beams=n_beams,
early_stopping=True
)
output=tokenizer.decode(beam_output[0], skip_special_tokens=True)
return output
inputs = [
gr.inputs.Textbox(label="prompt", lines=5, default="Bicalutamide"),
gr.Dropdown(model_names, value="BioGPT", label="selected_model"),
gr.inputs.Slider(1, 500, 1, default=100, label="min_len"),
gr.inputs.Slider(1, 2048, 1, default=1024, label="max_len"),
gr.inputs.Slider(1, 10, 1, default=5, label="num_beams")
]
outputs = gr.outputs.Textbox(label="output")
examples = [
["Bicalutamide", "BioGPT", 25, 100, 5],
["Janus kinase 3 (JAK-3)", "BioGPT", 25, 100, 5],
["Apricitabine", "BioGPT", 25, 100, 5],
]
iface = gr.Interface(
fn=get_beam_output,
inputs=inputs,
outputs=outputs,
examples=examples,
title="BioGPT: generative pre-trained transformer for biomedical text generation and mining"
)
iface.launch(debug=True, enable_queue=True)