Spaces:
Running
Running
File size: 2,554 Bytes
4bba8df 853f29a 4bba8df fb1a253 4bba8df fb1a253 4bba8df fb1a253 853f29a 4bba8df 2b41a25 4bba8df |
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 |
import gradio as gr
import os
import torch
import numpy as np
from transformers import AutoModelForSequenceClassification
from transformers import AutoTokenizer
from huggingface_hub import HfApi
from label_dicts import EMOTION9_LABEL_NAMES
from .utils import is_disk_full, release_model
HF_TOKEN = os.environ["hf_read"]
languages = [
"Czech", "English", "German", "Hungarian", "Polish", "Slovak"
]
domains = {
"parliamentary speech": "parlspeech",
}
def build_huggingface_path(language: str):
language = language.lower()
return f"poltextlab/xlm-roberta-large-pooled-{language}-emotions9"
def predict(text, model_id, tokenizer_id):
device = torch.device("cpu")
# Load JIT-traced model
jit_model_path = f"/data/jit_models/{model_id.replace('/', '_')}.pt"
model = torch.jit.load(jit_model_path).to(device)
model.eval()
# Load tokenizer (still regular HF)
tokenizer = AutoTokenizer.from_pretrained(tokenizer_id)
# Tokenize input
inputs = tokenizer(
text,
max_length=256,
truncation=True,
padding="do_not_pad",
return_tensors="pt"
)
inputs = {k: v.to(device) for k, v in inputs.items()}
with torch.no_grad():
output = model(inputs["input_ids"], inputs["attention_mask"])
print(output) # debug
logits = output["logits"]
release_model(model, model_id)
probs = torch.nn.functional.softmax(logits, dim=1).cpu().numpy().flatten()
NUMS_DICT = {i: key for i, key in enumerate(sorted(EMOTION9_LABEL_NAMES.keys()))}
output_pred = {f"[{NUMS_DICT[i]}] {EMOTION9_LABEL_NAMES[NUMS_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_e6(text, language, domain):
model_id = build_huggingface_path(language)
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="Emotions (9) Babel Demo",
fn=predict_e6,
inputs=[gr.Textbox(lines=6, label="Input"),
gr.Dropdown(languages, label="Language", value=languages[1]),
gr.Dropdown(domains.keys(), label="Domain", value=list(domains.keys())[0])],
outputs=[gr.Label(num_top_classes=5, label="Output"), gr.Markdown()])
|