📘 DocNLI-BERT (Finetuned for Natural Language Inference)
🧾 Model Overview
This model is a BERT-based sequence classification model finetuned on a Document NLI (Natural Language Inference) dataset.
Task:
Given a premise and a hypothesis, the model predicts whether the hypothesis is entailed by the premise or not.
- Base model:
bert-base-uncased
- Fine-tuned head: Sequence classification head with 2 labels (
entailment, not_entailment)
📊 Training Details
- Framework: Hugging Face Transformers + PEFT (LoRA)
- Optimizer: AdamW
- Learning rate: 5e-5
- Batch size: 16
- Epochs: 6
- Loss function: CrossEntropyLoss
- Best checkpoint selection: Based on validation accuracy
🚀 Usage
Load model & tokenizer
from transformers import AutoTokenizer, AutoModelForSequenceClassification
model_name = "ProdD/docnli-bert-finetuned"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)
