FP16xINT4 LLM inference kernel that can achieve near-ideal ~4x speedups up to medium batchsizes of 16-32 tokens.
☆1,018Sep 4, 2024Updated last year
Alternatives and similar repositories for marlin
Users that are interested in marlin are comparing it to the libraries listed below
Sorting:
- [MLSys'25] QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving; [MLSys'25] LServe: Efficient Long-sequence LLM Se…☆816Mar 6, 2025Updated 11 months ago
- FlashInfer: Kernel Library for LLM Serving☆5,009Updated this week
- BitBLAS is a library to support mixed-precision matrix multiplications, especially for quantized LLM deployment.☆751Aug 6, 2025Updated 6 months ago
- [MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration☆3,441Jul 17, 2025Updated 7 months ago
- Boosting 4-bit inference kernels with 2:4 Sparsity☆93Sep 4, 2024Updated last year
- [MLSys'24] Atom: Low-bit Quantization for Efficient and Accurate LLM Serving☆336Jul 2, 2024Updated last year
- QQQ is an innovative and hardware-optimized W4A8 quantization solution for LLMs.☆154Aug 21, 2025Updated 6 months ago
- Fast low-bit matmul kernels in Triton☆433Feb 1, 2026Updated 3 weeks ago
- A throughput-oriented high-performance serving framework for LLMs☆946Oct 29, 2025Updated 4 months ago
- Distributed Compiler based on Triton for Parallel Systems☆1,361Feb 13, 2026Updated 2 weeks ago
- Tile primitives for speedy kernels☆3,183Updated this week
- [ICML 2023] SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models☆1,612Jul 12, 2024Updated last year
- [ICML 2024] KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache☆358Nov 20, 2025Updated 3 months ago
- ☆261Jul 11, 2024Updated last year
- Code for the ICLR 2023 paper "GPTQ: Accurate Post-training Quantization of Generative Pretrained Transformers".☆2,258Mar 27, 2024Updated last year
- Code for Neurips24 paper: QuaRot, an end-to-end 4-bit inference of large language models.☆483Nov 26, 2024Updated last year
- AutoAWQ implements the AWQ algorithm for 4-bit quantization with a 2x speedup during inference. Documentation:☆2,314May 11, 2025Updated 9 months ago
- A fast communication-overlapping library for tensor/expert parallelism on GPUs.☆1,261Aug 28, 2025Updated 6 months ago
- LightLLM is a Python-based LLM (Large Language Model) inference and serving framework, notable for its lightweight design, easy scalabili…☆3,901Feb 20, 2026Updated last week
- GPTQ inference Triton kernel☆321May 18, 2023Updated 2 years ago
- Serving multiple LoRA finetuned LLM as one☆1,145May 8, 2024Updated last year
- [NeurIPS'24 Spotlight, ICLR'25, ICML'25] To speed up Long-context LLMs' inference, approximate and dynamic sparse calculate the attention…☆1,188Sep 30, 2025Updated 5 months ago
- [ICML 2024] Break the Sequential Dependency of LLM Inference Using Lookahead Decoding☆1,315Mar 6, 2025Updated 11 months ago
- Fast Matrix Multiplications for Lookup Table-Quantized LLMs☆388Apr 13, 2025Updated 10 months ago
- A Easy-to-understand TensorOp Matmul Tutorial☆410Feb 11, 2026Updated 2 weeks ago
- PyTorch native quantization and sparsity for training and inference☆2,696Updated this week
- An efficient GPU support for LLM inference with x-bit quantization (e.g. FP6,FP5).☆277Jul 16, 2025Updated 7 months ago
- [NeurIPS 2024] KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization☆402Aug 13, 2024Updated last year
- Transformers-compatible library for applying various compression algorithms to LLMs for optimized deployment with vLLM☆2,787Updated this week
- Official Implementation of EAGLE-1 (ICML'24), EAGLE-2 (EMNLP'24), and EAGLE-3 (NeurIPS'25).☆2,188Feb 20, 2026Updated last week
- Mirage Persistent Kernel: Compiling LLMs into a MegaKernel☆2,141Feb 19, 2026Updated last week
- Perplexity GPU Kernels☆564Nov 7, 2025Updated 3 months ago
- Official implementation of Half-Quadratic Quantization (HQQ)☆913Dec 18, 2025Updated 2 months ago
- ☆206May 5, 2025Updated 9 months ago
- A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit and 4-bit floating point (FP8 and FP4) precision on H…☆3,170Feb 21, 2026Updated last week
- High-speed GEMV kernels, at most 2.7x speedup compared to pytorch baseline.☆128Jul 13, 2024Updated last year
- A standalone GEMM kernel for fp16 activation and quantized weight, extracted from FasterTransformer☆96Feb 20, 2026Updated last week
- [ICML 2024] SqueezeLLM: Dense-and-Sparse Quantization☆713Aug 13, 2024Updated last year
- ☆580Oct 29, 2024Updated last year