Spaces:
Running
Running
File size: 3,280 Bytes
4bba8df 89d4ec8 4bba8df c259974 a27d7dd 89d4ec8 c259974 89d4ec8 c259974 89d4ec8 4bba8df 1849f87 4bba8df 3f9f187 4bba8df c259974 4bba8df 65e39ca |
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 96 97 98 99 100 101 102 |
import gradio as gr
import os
import torch
import numpy as np
import pandas as pd
from transformers import AutoModelForSequenceClassification
from transformers import AutoTokenizer
from huggingface_hub import HfApi
from label_dicts import CAP_MIN_NUM_DICT, CAP_MIN_LABEL_NAMES, CAP_LABEL_NAMES
from .utils import is_disk_full
HF_TOKEN = os.environ["hf_read"]
languages = [
"Multilingual",
]
domains = {
"media": "media",
"social media": "social",
"parliamentary speech": "parlspeech",
"legislative documents": "legislative",
"executive speech": "execspeech",
"executive order": "execorder",
"party programs": "party",
"judiciary": "judiciary",
"budget": "budget",
"public opinion": "publicopinion",
"local government agenda": "localgovernment"
}
def convert_minor_to_major(minor_topic):
if minor_topic == 999:
major_code = 999
else:
major_code = str(minor_topic)[:-2]
label = CAP_LABEL_NAMES[int(major_code)]
return label
def check_huggingface_path(checkpoint_path: str):
try:
hf_api = HfApi(token=HF_TOKEN)
hf_api.model_info(checkpoint_path, token=HF_TOKEN)
return True
except:
return False
def build_huggingface_path(language: str, domain: str):
return "poltextlab/xlm-roberta-large-pooled-cap-minor-v3"
def predict(text, model_id, tokenizer_id):
device = torch.device("cpu")
model = AutoModelForSequenceClassification.from_pretrained(model_id, low_cpu_mem_usage=True, device_map="auto", offload_folder="offload", token=HF_TOKEN)
tokenizer = AutoTokenizer.from_pretrained(tokenizer_id)
inputs = tokenizer(text,
max_length=256,
truncation=True,
padding="do_not_pad",
return_tensors="pt").to(device)
model.eval()
with torch.no_grad():
logits = model(**inputs).logits
probs = torch.nn.functional.softmax(logits, dim=1).cpu().numpy().flatten()
output_pred = {f"[{'999' if str(CAP_MIN_NUM_DICT[i]) == '999' else str(CAP_MIN_NUM_DICT[i])[:-2]}] {convert_minor_to_major(CAP_MIN_NUM_DICT[i])} - [{CAP_MIN_NUM_DICT[i]}] {CAP_MIN_LABEL_NAMES[CAP_MIN_NUM_DICT[i]]}": probs[i] for i in np.argsort(probs)[::-1]}
output_info = f'<p style="text-align: center; display: block">Prediction was made using the <a href="https://huggingface.co/{model_id}">{model_id}</a> model.</p>'
return output_pred, output_info
def predict_cap(text, language, domain):
domain = domains[domain]
model_id = build_huggingface_path(language, domain)
tokenizer_id = "xlm-roberta-large"
if is_disk_full():
os.system('rm -rf /data/models*')
os.system('rm -r ~/.cache/huggingface/hub')
return predict(text, model_id, tokenizer_id)
demo = gr.Interface(
title="CAP Minor Topics Babel Demo",
fn=predict_cap,
inputs=[gr.Textbox(lines=6, label="Input"),
gr.Dropdown(languages, label="Language"),
gr.Dropdown(domains.keys(), label="Domain")],
outputs=[gr.Textbox(num_top_classes=5, label="Output", elem_id="output-box"), gr.Markdown()],
css="""
#output-box {
width: 100% !important;
max-width: 800px;
}
""")
|