conftrace_

training-inference discrepancy

3 papers

Explore in graph

Also known as

TID

Co-occurring keywords · papers mentioning both

text generation (2) error accumulation (1) word-level training (1) transformer model (1) context sampling (1) self-generated text (1) probability scheduling (1) layer normalization (1) discrete diffusion model (1) neural machine translation (1)

Papers

Addressing the Training-Inference Discrepancy in Discrete Diffusion for Text Generation COLING 2025 Understanding the Failure of Batch Normalization for Transformers in NLP NeurIPS 2022 Bridging the Gap between Training and Inference for Neural Machine Translation ACL 2019