kwai / Megatron-KwaiLinks
[USENIX ATC '24] Accelerating the Training of Large Language Models using Efficient Activation Rematerialization and Optimal Hybrid Parallelism
☆61Updated last year
Alternatives and similar repositories for Megatron-Kwai
Users that are interested in Megatron-Kwai are comparing it to the libraries listed below
Sorting:
- ☆145Updated 5 months ago
- PyTorch bindings for CUTLASS grouped GEMM.☆135Updated last month
- nnScaler: Compiling DNN models for Parallel Training☆115Updated this week
- A baseline repository of Auto-Parallelism in Training Neural Networks☆144Updated 3 years ago
- ☆97Updated 11 months ago
- Train speculative decoding models effortlessly and port them smoothly to SGLang serving.☆313Updated this week
- ☆279Updated last week
- Sequence-level 1F1B schedule for LLMs.☆31Updated 2 months ago
- A lightweight design for computation-communication overlap.☆155Updated last week
- ☆113Updated 9 months ago
- ☆43Updated last year
- Zero Bubble Pipeline Parallelism☆418Updated 3 months ago
- High performance Transformer implementation in C++.☆129Updated 7 months ago
- PyTorch distributed training acceleration framework☆51Updated last week
- Pipeline Parallelism Emulation and Visualization☆60Updated 2 months ago
- Allow torch tensor memory to be released and resumed later☆109Updated last week
- A collection of memory efficient attention operators implemented in the Triton language.☆277Updated last year
- Dynamic Memory Management for Serving LLMs without PagedAttention☆407Updated 2 months ago
- Compare different hardware platforms via the Roofline Model for LLM inference tasks.☆111Updated last year
- Since the emergence of chatGPT in 2022, the acceleration of Large Language Model has become increasingly important. Here is a list of pap…☆266Updated 5 months ago
- ATC23 AE☆46Updated 2 years ago
- ☆98Updated 3 months ago
- flash attention tutorial written in python, triton, cuda, cutlass☆404Updated 3 months ago
- Flash-LLM: Enabling Cost-Effective and Highly-Efficient Large Generative Model Inference with Unstructured Sparsity☆217Updated last year
- ☆150Updated 7 months ago
- USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference☆549Updated last month
- A prefill & decode disaggregated LLM serving framework with shared GPU memory and fine-grained compute isolation.☆104Updated 3 months ago
- ☆128Updated 7 months ago
- Disaggregated serving system for Large Language Models (LLMs).☆669Updated 4 months ago
- A low-latency & high-throughput serving engine for LLMs☆405Updated 2 months ago