Add illframes to the demo

interfaces/illframes.py

@@ -0,0 +1,75 @@
+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 ILLFRAMES_MIGRATION_LABEL_NAMES, ILLFRAMES_COVID_LABEL_NAMES
+
+HF_TOKEN = os.environ["hf_read"]
+
+languages = [
+    "English"
+]
+
+domains = {
+    "Covid": "covid",
+    "Migration": "migration"
+}
+
+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(domain: str):
+    return f"poltextlab/xlm-roberta-large-english-ILLFRAMES-{domain}"
+
+def predict(text, model_id, tokenizer_id, label_names):
+    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()
+
+    NUMS_DICT = {i: key for i, key in enumerate(sorted(label_names.keys()))}
+
+    output_pred = {f"[{NUMS_DICT[i]}] {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_illframes(text, domain):
+    domain = domains[domain]
+    model_id = build_huggingface_path(domain)
+    tokenizer_id = "xlm-roberta-large"
+
+    if domain == "migration":
+        label_names = ILLFRAMES_MIGRATION_LABEL_NAMES
+    else:
+        label_names = ILLFRAMES_COVID_LABEL_NAMES
+
+    return predict(text, model_id, tokenizer_id, label_names)
+
+demo = gr.Interface(
+    fn=predict_cap,
+    inputs=[gr.Textbox(lines=6, label="Input"),
+            gr.Dropdown(languages, label="Language"),
+            gr.Dropdown(domains.keys(), label="Domain")],
+    outputs=[gr.Label(num_top_classes=5, label="Output"), gr.Markdown()])