Liger-Kernel
Liger-Kernel is a collection of efficient Triton cores designed for large language model (LLM) instruction. This library, which improves multiple GPU training efficiency and reduces memory usage, offers components compatible with Hugging Face. Flash Attention can work directly with PyTorch FSDP and Microsoft DeepSpeed.
linkedin · Local AI / Self-Hosted LLM
Common commercial products in this category: ChatGPT, Claude, Gemini
- You use Triton kernels such as the Hugging Face compatible RMSNorm, RoPE, SwiGLU and CrossEntropy.
- You save memory with lost functions like DPO, CPO, ORPO, and SimPO.
- You integrate with Flash Attention, PyTorch FSDP and Microsoft DeepSpeed.
- ↑ PositiveHacker News
Liger-Kernel: Efficient Triton kernels for LLM training
“Maximizing GPU efficiency for training large language models (LLMs) is challenging due to issues like out-of-memory (OOM) errors, scaling batch size, and sequence length. To address these challenges, LinkedIn has developed an open-source library called Liger-Kernel, which offers…”
- • NeutralHacker News
Liger-Kernel: Efficient Triton kernels for LLM training
“Liger-Kernel support has already been merged to axolotl [0] along with an example config that makes use of it [1], if anyone would like to quickly try it out. [0] [1]”