Cross Lingual Natural Language Inference With Entailment Scoring

via “entailment score interpretation and confidence ranking”

zero-shot-classification model by undefined. 26,55,180 downloads.

Unique: Exposes three-way entailment judgments rather than binary classification, providing richer confidence signals and enabling neutral-class-based uncertainty detection

vs others: More interpretable than softmax-only classifiers due to explicit entailment reasoning; attention visualization more meaningful than black-box confidence scores

via “cross-lingual-natural-language-inference”

zero-shot-classification model by undefined. 3,03,704 downloads.

Unique: Trained on XNLI's 2.7M examples across 15 languages with DeBERTa-v3's disentangled attention, which explicitly separates content and position information in attention heads. This architectural choice allows the model to learn language-agnostic entailment patterns that transfer across typologically distant languages (e.g., English to Japanese) better than standard BERT-style models.

vs others: Achieves 85%+ accuracy on XNLI benchmark vs 75-80% for XLM-RoBERTa, and unlike task-specific models (e.g., RoBERTa-large-mnli), maintains strong cross-lingual transfer without requiring language-specific fine-tuning.

via “cross-lingual natural language inference with entailment scoring”

zero-shot-classification model by undefined. 2,28,003 downloads.

Unique: Trained jointly on MNLI (English, 433K examples) and XNLI (15 languages, 75K examples), enabling zero-shot cross-lingual entailment without language-specific fine-tuning. DeBERTa-v3's disentangled attention mechanism explicitly separates content and position information, improving cross-lingual generalization compared to standard transformer architectures.

vs others: Achieves 2-5% higher accuracy on XNLI multilingual benchmarks than mBERT and XLM-R due to DeBERTa's attention design, and requires no language-specific adapters unlike adapter-based approaches, making it faster to deploy across new languages.

via “multi-dataset-nli-entailment-scoring”

zero-shot-classification model by undefined. 2,25,548 downloads.

Unique: Trained on FEVER (fact-checking claims), ANLI (adversarial NLI), and WANLI (weak supervision) in addition to standard MNLI, capturing adversarial examples and noisy labels that improve robustness to edge cases and adversarial inputs compared to single-dataset NLI models

vs others: More robust to adversarial premise-hypothesis pairs than MNLI-only models; FEVER training improves fact-checking accuracy by 3-5% on out-of-domain claims vs. RoBERTa-MNLI baselines

via “natural language inference scoring for semantic entailment”

zero-shot-classification model by undefined. 1,46,288 downloads.

Unique: Fine-tuned on XNLI (cross-lingual NLI) dataset covering 15 languages, enabling entailment scoring that works across languages without language-specific NLI models, using a shared 3-class head (entailment/contradiction/neutral) rather than task-specific classifiers

vs others: Provides language-agnostic entailment scoring vs monolingual NLI models, and enables zero-shot classification via NLI reformulation unlike traditional classifiers that require labeled data per task

via “semantic entailment scoring for ranking and retrieval”

zero-shot-classification model by undefined. 1,87,439 downloads.

Unique: Provides direct entailment classification rather than embedding-based similarity, enabling explicit logical relationship scoring. The cross-encoder architecture ensures that entailment scores reflect the joint context of both premise and hypothesis, unlike bi-encoder approaches that score embeddings independently.

vs others: More semantically precise than embedding-based ranking (e.g., sentence-transformers bi-encoders) for entailment-specific tasks because it directly models logical relationships, though slower due to cross-encoder architecture; better for fact-checking and QA ranking, worse for large-scale retrieval due to latency.

via “zero-shot natural language inference classification”

zero-shot-classification model by undefined. 2,58,745 downloads.

Unique: Uses a distilled cross-encoder architecture (MiniLMv2-L6-H768, 22.7M parameters) that jointly encodes premise-hypothesis pairs through a single transformer pass, enabling direct interaction modeling while maintaining <100ms inference latency on CPU — a balance point between bi-encoder speed and cross-encoder accuracy that most alternatives sacrifice

vs others: Faster than full-size cross-encoder NLI models (RoBERTa-Large) by 3-5x due to distillation, yet maintains competitive zero-shot entailment accuracy; slower than bi-encoder alternatives for ranking but captures semantic interactions that bi-encoders miss

via “sentence-pair entailment scoring with probability calibration”

zero-shot-classification model by undefined. 2,47,798 downloads.

Unique: Provides calibrated probability distributions trained jointly on SNLI (570K pairs) and MultiNLI (433K pairs) using cross-entropy loss, enabling direct use of softmax outputs for confidence-based filtering without additional calibration layers, unlike single-dataset models that often require temperature scaling

vs others: More calibrated than zero-shot LLM-based NLI (which often produce overconfident probabilities) and faster than ensemble approaches, while maintaining comparable accuracy to larger models like DeBERTa-base

via “premise-hypothesis entailment scoring for classification”

zero-shot-classification model by undefined. 1,17,720 downloads.

Unique: Reformulates classification as NLI by treating category labels as hypotheses and computing entailment scores, enabling zero-shot inference without task-specific training. This approach leverages the model's NLI pretraining to generalize to arbitrary categories defined at inference time.

vs others: Entailment-based classification outperforms simple semantic similarity approaches (e.g., embedding cosine distance) by 5-10% on zero-shot tasks because it explicitly models logical relationships rather than just semantic proximity.

via “multi-dataset natural language inference with cross-domain robustness”

zero-shot-classification model by undefined. 64,968 downloads.

Unique: Combines three complementary NLI datasets (MNLI for general inference, FEVER for fact-checking, ANLI for adversarial robustness) with DeBERTa-v3's disentangled attention to create a model that generalizes across domains and resists adversarial examples; adversarial training on ANLI specifically targets common NLI failure modes

vs others: More robust to adversarial and out-of-domain examples than single-dataset NLI models (e.g., MNLI-only BERT) due to multi-dataset training; smaller and faster than T5-based NLI models while maintaining competitive accuracy on FEVER and ANLI benchmarks

via “semantic similarity scoring via entailment logits”

text-classification model by undefined. 5,13,435 downloads.

Unique: Repurposes entailment logits as a similarity proxy without explicit fine-tuning on similarity tasks. The disentangled attention mechanism enables the model to capture both semantic and structural relationships, making entailment-based similarity more nuanced than simple cosine similarity on embeddings. However, this approach is fundamentally indirect and requires careful calibration.

vs others: Faster than dedicated similarity models (e.g., Sentence-BERT) because it reuses the same model for both inference and similarity; more interpretable than embedding-based similarity because entailment logits provide explicit reasoning signals (entailment vs. contradiction vs. neutral).

via “multilingual natural language inference with english-primary training”

zero-shot-classification model by undefined. 33,943 downloads.

Unique: Combines four diverse NLI training datasets (MNLI for formal reasoning, FEVER for factual claims, ANLI for adversarial robustness, LingNLI for linguistic phenomena) into a single model checkpoint, leveraging DeBERTa-v3's disentangled attention to learn dataset-specific reasoning patterns while maintaining generalization; binary variant simplifies deployment for entailment-only use cases

vs others: Achieves higher accuracy on out-of-domain NLI benchmarks than RoBERTa-large-mnli and ELECTRA-large-discriminator while using 7x fewer parameters, and the multi-dataset training provides better robustness to adversarial examples and factual claims compared to single-dataset MNLI-only models

via “natural language inference with sentence-pair classification”

* 🏆 2020: [Language Models are Few-Shot Learners (GPT-3)](https://proceedings.neurips.cc/paper/2020/hash/1457c0d6bfcb4967418bfb8ac142f64a-Abstract.html)

Unique: Leverages the [CLS] token representation (pre-trained via NSP objective) for sentence-pair classification, creating a direct connection between pre-training and fine-tuning objectives; bidirectional context enables understanding of semantic relationships without explicit alignment or interaction mechanisms

vs others: Achieves +4.6 percentage point improvement on MultiNLI compared to prior baselines by using bidirectional context and joint pre-training (MLM + NSP), whereas prior approaches required task-specific interaction layers or attention mechanisms