NVIDIA / TransformerEngine
A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit floating point (FP8) precision on Hopper, Ada and Blackwell GPUs, to provide better performance with lower memory utilization in both training and inference.
☆2,365Updated this week
Alternatives and similar repositories for TransformerEngine:
Users that are interested in TransformerEngine are comparing it to the libraries listed below
- FlashInfer: Kernel Library for LLM Serving☆2,693Updated this week
- [ICML 2023] SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models☆1,392Updated 9 months ago
- [MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration☆2,942Updated last week
- PyTorch native quantization and sparsity for training and inference☆1,974Updated this week
- Pipeline Parallelism for PyTorch☆764Updated 8 months ago
- Ongoing research training transformer language models at scale, including: BERT & GPT-2☆2,049Updated last month
- Transformer related optimization, including BERT, GPT☆6,137Updated last year
- nvidia-modelopt is a unified library of state-of-the-art model optimization techniques like quantization, pruning, distillation, speculat…☆870Updated last week
- PyTorch extensions for high performance and large scale training.☆3,306Updated 2 weeks ago
- Code for the ICLR 2023 paper "GPTQ: Accurate Post-training Quantization of Generative Pretrained Transformers".☆2,093Updated last year
- Kernl lets you run PyTorch transformer models several times faster on GPU with a single line of code, and is designed to be easily hackab…☆1,565Updated last year
- FP16xINT4 LLM inference kernel that can achieve near-ideal ~4x speedups up to medium batchsizes of 16-32 tokens.☆803Updated 7 months ago
- MII makes low-latency and high-throughput inference possible, powered by DeepSpeed.☆2,002Updated 3 weeks ago
- AutoAWQ implements the AWQ algorithm for 4-bit quantization with a 2x speedup during inference. Documentation:☆2,104Updated 2 weeks ago
- The Triton TensorRT-LLM Backend☆827Updated last week
- [ICML 2024] Break the Sequential Dependency of LLM Inference Using Lookahead Decoding☆1,242Updated last month
- A Python-level JIT compiler designed to make unmodified PyTorch programs faster.☆1,040Updated last year
- Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels☆1,052Updated this week
- A pytorch quantization backend for optimum☆922Updated last week
- Minimalistic large language model 3D-parallelism training☆1,793Updated this week
- Microsoft Automatic Mixed Precision Library☆593Updated 6 months ago
- A throughput-oriented high-performance serving framework for LLMs☆797Updated 7 months ago
- Automatically Discovering Fast Parallelization Strategies for Distributed Deep Neural Network Training☆1,783Updated 3 weeks ago
- Tutel MoE: Optimized Mixture-of-Experts Library, Support DeepSeek FP8/FP4☆803Updated this week
- Tile primitives for speedy kernels☆2,279Updated this week
- Puzzles for learning Triton☆1,591Updated 5 months ago
- Transformers-compatible library for applying various compression algorithms to LLMs for optimized deployment with vLLM☆1,251Updated this week
- A PyTorch Native LLM Training Framework☆792Updated 3 months ago
- Mirage: Automatically Generating Fast GPU Kernels without Programming in Triton/CUDA☆808Updated this week
- Flash Attention in ~100 lines of CUDA (forward pass only)☆786Updated 3 months ago