Initial commit

.gitignore

@@ -0,0 +1,2 @@
+.venv
+.python-version

app.py

@@ -0,0 +1,28 @@
+import gradio
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+from scipy.special import softmax
+
+
+def predict(text):
+    model_name = "deepset/bert-base-german-cased-hatespeech-GermEval18Coarse"
+    short_score_descriptions = {0: "Kein Hasskommentar", 1: "Hasskommentar"}
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForSequenceClassification.from_pretrained(model_name)
+    model_input = tokenizer(*([text],), padding=True, return_tensors="pt")
+    with torch.no_grad():
+        output = model(**model_input)
+        logits = softmax(output[0][0].detach().numpy()).tolist()
+    return {short_score_descriptions[k]: v for k, v in enumerate(logits)}
+
+
+gradio.Interface(
+    title="Klassifikator deutschsprachiger Hasskommentare",
+    inputs=[
+        gradio.Textbox(label="Kommentar"),
+    ],
+    fn=predict,
+    outputs=[
+        gradio.Label(label="Klassifikation"),
+    ],
+).launch()

requirements.txt

@@ -0,0 +1,4 @@
+torch==2.4.1
+gradio==4.44.0
+transformers==4.44.2
+scipy==1.14.1