Ascend / AscendSpeed
☆78Updated last year
Alternatives and similar repositories for AscendSpeed:
Users that are interested in AscendSpeed are comparing it to the libraries listed below
- ☆127Updated 3 months ago
- PyTorch bindings for CUTLASS grouped GEMM.☆108Updated 2 months ago
- Transformer related optimization, including BERT, GPT☆39Updated 2 years ago
- Compare different hardware platforms via the Roofline Model for LLM inference tasks.☆93Updated last year
- Transformer related optimization, including BERT, GPT☆59Updated last year
- Models and examples built with OneFlow☆96Updated 5 months ago
- ☆139Updated 11 months ago
- Odysseus: Playground of LLM Sequence Parallelism☆66Updated 9 months ago
- ☆125Updated 2 weeks ago
- Summary of system papers/frameworks/codes/tools on training or serving large model☆56Updated last year
- A collection of memory efficient attention operators implemented in the Triton language.☆253Updated 9 months ago
- An easy-to-use package for implementing SmoothQuant for LLMs☆95Updated 10 months ago
- FlagScale is a large model toolkit based on open-sourced projects.☆253Updated this week
- A MoE impl for PyTorch, [ATC'23] SmartMoE☆61Updated last year
- ☆74Updated 3 months ago
- Transformer related optimization, including BERT, GPT☆17Updated last year
- Train llm (bloom, llama, baichuan2-7b, chatglm3-6b) with deepspeed pipeline mode. Faster than zero/zero++/fsdp.☆94Updated last year
- [USENIX ATC '24] Accelerating the Training of Large Language Models using Efficient Activation Rematerialization and Optimal Hybrid Paral…☆51Updated 7 months ago
- ☆44Updated this week
- ☆15Updated 11 months ago
- ☆88Updated 6 months ago
- DashInfer is a native LLM inference engine aiming to deliver industry-leading performance atop various hardware architectures, including …☆237Updated 2 weeks ago
- ☆214Updated last year
- Efficient, Flexible, and Highly Fault-Tolerant Model Service Management Based on SGLang☆42Updated 4 months ago
- Fast and memory-efficient exact attention☆52Updated this week
- [ICLR 2025] PEARL: parallel speculative decoding with adaptive draft length☆59Updated last week
- USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference☆452Updated this week
- Sequence-level 1F1B schedule for LLMs.☆17Updated 9 months ago
- Simple Dynamic Batching Inference☆145Updated 3 years ago
- ☆23Updated last year