TinyLlama-augesc-context-strategy

Test set performance

• Top 1 Accuracy: 0.4346
• Top 3 Accuracy: 0.7677
• Top 1 Macro F1: 0.2668
• Top 3 Macro F1: 0.5669

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification

device="cuda:0"
model = "heegyu/TinyLlama-augesc-context-strategy"
tokenizer = AutoTokenizer.from_pretrained(model)
model = AutoModelForSequenceClassification.from_pretrained(model).eval().to(device)

example = """usr: Hi
sys[Question]: Hello, how are you today?
usr: I was scolded by my parents yesterday"""

inputs = tokenizer(example, return_tensors="pt").to(device)
logits = model(**inputs).logits.softmax(-1)
print(logits)

label = logits.argmax(-1).item()


ESCONV_STRATEGY = [
    "Question",
    "Restatement or Paraphrasing",
    "Reflection of feelings",
    "Self-disclosure",
    "Affirmation and Reassurance",
    "Providing Suggestions",
    "Information",
    "Others"
]
id2label = {i:k for i, k in enumerate(ESCONV_STRATEGY)}

print(id2label[label])