deberta-v3-large_boolq

This model is a fine-tuned version of microsoft/deberta-v3-large on the boolq dataset. It achieves the following results on the evaluation set:

• Loss: 0.4601
• Accuracy: 0.8835

Model description

More information needed

Example

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained("nfliu/deberta-v3-large_boolq")
tokenizer = AutoTokenizer.from_pretrained("nfliu/deberta-v3-large_boolq")

# Each example is a (question, context) pair.
examples = [
    ("Lake Tahoe is in California", "Lake Tahoe is a popular tourist spot in California."),
    ("Water is wet", "Contrary to popular belief, water is not wet.")
]

encoded_input = tokenizer(examples, padding=True, truncation=True, return_tensors="pt")

with torch.no_grad():
    model_output = model(**encoded_input)
    probabilities = torch.softmax(model_output.logits, dim=-1).cpu().tolist()

probability_no = [round(prob[0], 2) for prob in probabilities]
probability_yes = [round(prob[1], 2) for prob in probabilities]

for example, p_no, p_yes in zip(examples, probability_no, probability_yes):
    print(f"Question: {example[0]}")
    print(f"Context: {example[1]}")
    print(f"p(No | question, context): {p_no}")
    print(f"p(Yes | question, context): {p_yes}")
    print()

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.85	250	0.5306	0.8823
0.1151	1.69	500	0.4601	0.8835
0.1151	2.54	750	0.5897	0.8792
0.0656	3.39	1000	0.6477	0.8804
0.0656	4.24	1250	0.6847	0.8838

Framework versions

• Transformers 4.32.1
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.13.3