Wan2.2: Open and Advanced Large-Scale Video Generative Model
Technical principles related to large models
Deep learning optimization library: makes distributed training easy
Lets make video diffusion practical
Neural Network Compression Framework for enhanced OpenVINO
Unified KV Cache Compression Methods for Auto-Regressive Models
Redundancy-aware KV Cache Compression for Reasoning Models
SOTA discrete acoustic codec models with 40/75 tokens per second
Implementation of TurboQuant (ICLR 2026)
The highest-scoring AI memory system ever benchmarked
Awesome multilingual OCR toolkits based on PaddlePaddle
14-stage Fusion Pipeline for LLM token compression
A tension reasoning engine over 131 S-class problems
Data Lake for Deep Learning. Build, manage, and query datasets
Data and tools for generating and inspecting OLMo pre-training data
AIMET is a library that provides advanced quantization and compression
Libraries for applying sparsification recipes to neural networks
LMDeploy is a toolkit for compressing, deploying, and serving LLMs
Contexts Optical Compression
From-scratch PyTorch implementation of Google's TurboQuant
Compress tool outputs, logs, files, and RAG chunks
Trainable latent-memory framework for 100M-token contexts
A unified library of SOTA model optimization techniques
The official repository for ERNIE 4.5 and ERNIEKit
PaddlePaddle End-to-End Development Toolkit