deberta

Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"

Unattended Lightweight Text Classifiers with LLM Embeddings

🥇Samsung AI Challenge 2021 1등 솔루션입니다🥇

[ACL 2023] Solving Math Word Problems via Cooperative Reasoning induced Language Models (LLMs + MCTS + Self-Improvement)

Our EMNLP 2022 paper on MCQA

금융 도메인에 특화된 한국어 임베딩 모델

Premove ITN: context-aware inverse text normalization for English voice-agent transcripts.

Lightweight prompt-injection detector built for AI agents, copilots, and LLM apps. Fast, local, calibrated.

Local, per-sentence AI-text detection with pattern diagnostics — CLI + MCP server (DeBERTa-v3, RAID #1).

The implementation of DeBERTaV3-based commonsense question answering on CommonsenseQA.

Benchmarked comparison of multiple serving patterns for Transformer encoder models (BERT-style classification, NLI) on Amazon SageMaker AI — ONNX CPU, multi-container, inference component, LMI, scale-to-zero

Find exactly which parts of your LLM output are hallucinated. Claim-level factuality verification with NLI, calibrated confidence, and pipeline diagnostics for RAG systems.

Application for training the pretrained transformer model DeBERTaV3 on an Aspect Based Sentiment Analysis task

The code of Hierarchical Multi-label Classification (HMC). It is a final course project of Natural Language Processing and Deep Learning, 2022 Fall.

Deberta implementation at Cross-Domain Sentiment Classification With Bidirectional Contextualized Transformer Language Models

BERT agent with ERE P1-P5 verification gates

An 184M-parameter decision engine delivering calibrated distributions from one 22ms forward pass while matching models 2.3x its size.

71M parameter safety classifier (DeBERTa-v3-xsmall). Dual-head: binary safe/unsafe + 7 category labels. Sub-2ms inference.

Code release for KCLarity at SemEval-2026 Task 6: Encoder and Zero-Shot Approaches to Political Evasion Detection

Real-time LLM hallucination guardrail — NLI + RAG fact-checking with opt-in claim-level streaming contradiction halt. Drop-in for any LLM backend.