banking77

NIRNAY 450M: open-weight calibrated decision model (Apache-2.0). Banking77 intent classification 87.9%, fitted ECE 0.045. Laya fork plus concept bottleneck, coarse-to-fine pointer and RLCD. Built by Eulogik.

Banking intent classification using an MLP baseline and parameter-efficient LoRA fine-tuning of RoBERTa.

LLM drift & quality monitoring: embedding drift, output-quality drift, and LLM-as-judge regression tracking

Empirical study of candidate-order instability in non-autoregressive multi-candidate transformers, with cardinality scaling, permutation marginalization, cyclic shifts, calibration, and reproducible statistical evaluation.

Verifiable handoffs between AI systems: Waggle packages decisions and evidence; Kea independently verifies, replays, or refuses them.

Held-out intent benchmark on BANKING77 for few-shot intent classification, comparing TF-IDF, frozen MiniLM, and ProtoNet on unseen intents.

Small example of fine tuning ettin-encoder-17m over banking77 dataset on CPU only

Fine-tuned DeBERTa-v3-base for 77-class BANKING77 intent classification with leakage-aware evaluation and statistical comparison against a TF-IDF baseline.

Классификация банковских обращений: TF-IDF и BERT Tiny, оценка уверенности и демо на Streamlit.

Benchmark and results for Closing the Compliance Gap: open-weights and proprietary LLMs for privacy-constrained contact center automation (under review, NeurIPS 2026 InfPriv workshop)

Fine-tunes Google FLAN-T5-base with LoRA (PEFT) on Banking77 to classify 77 banking intents. Trains only 1.77M params (0.71%) in ~2hrs on free Colab GPU. Served via FastAPI REST API with batch prediction, Docker support & evaluation metrics.

Reproducible Laya-CoreML vs Jev benchmark for zero-shot intent classification on Banking77, ArBanking77, and CLINC150. Includes paired accuracy and per-dataset results.

NLP lab comparing a Hybrid XGBoost against a fine-tuned DistilBERT for 77-way banking intent classification, evaluating F1, latency and structural trade-offs.

Which bank support tickets can a machine safely handle, which must reach a person, and what each approach costs. Benchmarks keyword rules, classic ML and Gemini on BANKING77, scored by harm, not just accuracy.

Frozen encoder + Mahalanobis prototype for class-incremental intent classification. 50+ experiments across BANKING77, CLINC150, HWU64, AG News. Matches fine-tuned baselines at 5MB state with zero forgetting, order-invariance, 455 QPS.

Local Jev-style System One decision models on Apple Silicon Macs: AnyJev, Kev, Laya (Ollaya) and CLM benchmarked on English, Korean and BANKING77 (accuracy, latency, order-flip)

Measured experiments on decision models. Can an open-weight model re-rank search results? Does a token budget explain a benchmark failure? What does one typed decision cost across model families? Every request and response is recorded, and negative results are published as they came out.

A framework for benchmarking model reliability beyond clean accuracy using stability, robustness, severity-response, and paired evaluation.

BANKING77 intent classification: LoRA vs TF-IDF, held-out evaluation, and a guarded local API