Intel® Gaudi® AI Accelerator

Model-References

Reference models for Intel(R) Gaudi(R) AI Accelerator

Gaudi-tutorials

Tutorials for running models on First-gen Gaudi and Gaudi2 for Training and Inference. The source files for the tutorials on https://developer.habana.ai/

SynapseAI_Core

SynapseAI Core is a reference implementation of the SynapseAI API running on Habana Gaudi

vllm-fork

A high-throughput and memory-efficient inference and serving engine for LLMs

Setup_and_Install

Setup and Installation Instructions for Habana binaries, docker image creation

Habana_Custom_Kernel

Provides the examples to write and build Habana custom kernels using the HabanaTools

hccl_demo

DeepSpeed

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

Gaudi-solutions

Full End-to-End examples showing how to use First-gen Gaudi and Gaudi2 in common use cases

Gaudi2-Workshop

deepspeed_old

hl-thunk-open

Thunk library for HabanaLabs kernel driver

Megatron-DeepSpeed

Intel Gaudi's Megatron DeepSpeed Large Language Models for training

habana-container-runtime

Habana container runtime

gohlml

HABANA Management Library bindings for Go

habanalabs-k8s-device-plugin

HABANA device plugin for Kubernetes

Fairseq

optimum-habana-fork

Easy and lightning fast training of 🤗 Transformers on Habana Gaudi processor (HPU)

pytorch-lightning

The lightweight PyTorch wrapper for high-performance AI research. Scale your models, not the boilerplate.

DL1-Workshop

drivers.accel.habanalabs.kernel

hccl_ofi_wrapper

papers

Academic papers by Habana research team

slurm

Slurm: A Highly Scalable Workload Manager

AutoGPTQ

An easy-to-use LLMs quantization package with user-friendly apis, based on GPTQ algorithm.

DeepSpeedExamples

Example models using DeepSpeed

drivers.gpu.linux-nic.kernel

NIC drivers (Ethernet, IBverbs and common) for the NIC IP that is inside Intel's data-center GPU

Intel_Gaudi3_Software

Intel® Gaudi® Software is an implementation of the runtime and graph compiler for Gaudi3

neural-compressor

SOTA low-bit LLM quantization (INT8/FP8/INT4/FP4/NF4) & sparsity; leading model compression techniques on TensorFlow, PyTorch, and ONNX Runtime

rdma-core

RDMA core userspace libraries and daemons