Search Results for "decoder software"

Jacobian Lens is Anthropic’s reference implementation for examining what a language model’s internal activations are inclined to produce as text. It transports residual-stream vectors from selected layers and positions into the final-layer basis using an averaged input-output Jacobian. The transformed vectors are decoded through the model’s own unembedding into ranked vocabulary predictions. The package can fit new lenses, load saved ones, apply them to prompts, and merge results from...

This project, also known as TorchMultimodal, is a PyTorch library for building, training, and experimenting with multimodal, multi-task models at scale. The library provides modular building blocks such as encoders, fusion modules, loss functions, and transformations that support combining modalities (vision, text, audio, etc.) in unified architectures. It includes a collection of ready model classes—like ALBEF, CLIP, BLIP-2, COCA, FLAVA, MDETR, and Omnivore—that serve as reference...

DeepLabv3 Plus is a PyTorch implementation of DeepLabv3+ for semantic segmentation. It implements the encoder-decoder architecture with atrous separable convolution and provides a practical workflow for training, prediction, and mIoU evaluation. The repository supports VOC-style segmentation datasets and includes utilities for annotation generation, JSON dataset conversion, model summary inspection, prediction, and metric calculation. It provides pretrained weight workflows for MobileNetV2...

PyTorch training code and pretrained models for DETR (DEtection TRansformer). We replace the full complex hand-crafted object detection pipeline with a Transformer, and match Faster R-CNN with a ResNet-50, obtaining 42 AP on COCO using half the computation power (FLOPs) and the same number of parameters. Inference in 50 lines of PyTorch. What it is. Unlike traditional computer vision techniques, DETR approaches object detection as a direct set prediction problem. It consists of a set-based...

seq2seq is an early, influential TensorFlow reference implementation for sequence-to-sequence learning with attention, covering tasks like neural machine translation, summarization, and dialogue. It packaged encoders, decoders, attention mechanisms, and beam search into a modular training and inference framework. The codebase showcased best practices for batching, bucketing by sequence length, and handling variable-length sequences efficiently on GPUs. Researchers used it as a baseline to...

BTK contains C++ and Python libraries that implement speech processing and microphone array techniques such as speech feature extraction, speech enhancement, speaker tracking, beamforming, dereverberation and echo cancellation algorithms. The Millennium ASR provides C++ and python libraries for automatic speech recognition. The Millennium ASR implements a weighted finite state transducer (WFST) decoder, training and adaptation methods. These toolkits are meant for facilitating research and...