Tech Times

Baidu OCR Breaks Long-Document Memory Wall: New Architecture Beats DeepSeek

Open-source OCR from Baidu eliminates the GPU memory wall that limits long-document parsing. Unlimited OCR uses a constant KV ...
IEEE

Interpretable Merging-Vehicle Acceleration Prediction via a Decoder-Only Transformer and Attention Map Clustering

Abstract: This paper proposes a decoder-only Transformer model for analyzing and clustering vehicle-to-vehicle interactions in highway merge zones. Leveraging deep learning, the model captures complex ...
GitHub

Input Encoders for Multi-Channel Signal Transformers

Recommendation. Default to nn.Linear(C, d_model) — it's the simplest thing that works and we never decisively beat it, because the per-channel projections spontaneously orthogonalise when the task ...
GitHub

GitHub - OssiLehtinen/channel-encoder-audit: Empirical audit of input encoders for multi-channel signal transformers — nn.Linear (C, d_model) matches every alternative. · GitHub

Recommendation. Default to nn.Linear (C, d_model) — it's the simplest thing that works and we never decisively beat it, because the per-channel projections spontaneously orthogonalise when the task ...
IEEE

Efficient Pruning and Acceleration of Encoder-Based LLM Transformers on eFPGAs

Abstract: Transformer encoders such as Bidirectional Encoder Representations from Transformers (BERT) are widely adopted for Natural Language Processing (NLP) tasks, yet their computational and memory ...