Showing content from https://github.com/sgl-project/ome/releases/latest below:
Release OME v.0.1.3 ยท sgl-project/ome ยท GitHub
๐ What's New Kimi K2 Model & Runtime Support
Full support has been added for the Kimi K2 model family, including model parsing and one of the most advanced runtime configurations to date. This release integrates the latest versions of SGLang and SGLang Router, with support for prefill and decode disaggregation, EPLB, and DeepEP.
Model Architecture Expansion
Added support for over 20 new model architectures in the Hugging Face model loader, significantly expanding the range of models that can be deployed and managed.
Multi-Cloud & CoreWeave Support
- Introduced support for storing benchmark results across multiple cloud storage providers.
- Added native support for CoreWeave infrastructure, including H200 instance shape configurations.
- Refactored node shape utilities to support instance types across multiple cloud providers.
Runtime Selection and CRD Enhancements
Serving runtimes now support weighted selection logic based on model format and framework version. These enhancements make the runtime selection process more intelligent and adaptable across diverse model types.
Model Replication & Authentication
Introduced first-class support for replicating models directly from Hugging Face to OCI Object Storage. This includes robust handling of authentication and authorization using both Hugging Face tokens and OCI IAM credentials.
๐ฆ Container Images
The following container images are available:
# OME Manager
docker pull ghcr.io/moirai-internal/ome-manager:v0.1.3
# Model Agent
docker pull ghcr.io/moirai-internal/model-agent:v0.1.3
# OME Agent
docker pull ghcr.io/moirai-internal/ome-agent:v0.1.3
# Multinode Prober
docker pull ghcr.io/moirai-internal/multinode-prober:v0.1.3
โ Helm Installation Option 1: OCI Registry (Recommended)
# Install directly from OCI registry
helm install ome-crd oci://ghcr.io/moirai-internal/charts/ome-crd --version 0.1.3 --namespace ome --create-namespace
helm install ome oci://ghcr.io/moirai-internal/charts/ome-resources --version 0.1.3 --namespace ome
Option 2: GitHub Releases
# Add the OME Helm repository
helm repo add ome https://github.com/sgl-project/ome/releases/download/v0.1.3
helm repo update
# Install OME
helm install ome-crd ome/ome-crd --namespace ome --create-namespace
helm install ome ome/ome-resources --namespace ome
New Contributors
What's Changed
- Renames SRT to SGLang for clarity by @JustinTong0323 in #125
- [BUG] Fix Inference Service Status Propagation by @YouNeedCryDear in #89
- [bugfix] fix: resolve model agent download loops and timeout issues by @slin1237 in #126
- [core] feat: add multi-cloud storage support for benchmark results by @slin1237 in #127
- [misc] Add support for 20+ new model architectures to HuggingFace modโฆ by @slin1237 in #128
- feat(helm): Add imagePullSecrets support to ome-resources chart by @abatilo in #131
- [docs] update intallation document by @slin1237 in #130
- Fix YAML syntax error in helm workflow by @abatilo in #133
- Fix typo: rename Scount to Scout by @abatilo in #135
- fix: use non-deprecated node instance type label by @abatilo in #132
- Fix model-agent health check timing to account for startup jitter by @abatilo in #136
- Refactor node shape utilities to support multi-cloud instance types by @abatilo in #134
- Add LWS worker-index 0 to sample DS manifest by @abatilo in #138
- [bugfix] fix nodeshape for unknown vendors by @slin1237 in #141
- [core] adds model node selector to components by @slin1237 in #142
- [bugfix] fix ingress annotation reconciliation by @slin1237 in #144
- [bugfix] Separate OCI Object Storage clients for ome agent by @chengjieyao in #143
- [misc] refactor component cleanup by @slin1237 in #145
- [docs] remove empty pages and fixed styling by @slin1237 in #146
- [docs] fix prism style issue in docs page by @slin1237 in #147
- [Bugfix]
make install
reports an erros : /bin/sh: 1: read: arg count by @mupeifeiyi in #150
- [Bugfix] Fix models yaml file no such file or directory by @mupeifeiyi in #152
- [DOC]Fix doc architecture diagram SVG title by @YouNeedCryDear in #154
- add modelFormat, modelFramework version compare by @pallasathena92 in #148
- [Misc] replica ome-agent: add support to replicate models from HF to OCI by @beiguo218 in #157
- [Misc] replica ome-agent: minor improvements + cleanup by @beiguo218 in #158
- [Bugfix] try to get baseModelName and protocolVersion from Model and โฆ by @leprode in #156
- [oep] add pvc storage support by @slin1237 in #169
- [core] add pvc storage parsing supporting base model and cluster baseโฆ by @slin1237 in #170
- [core] skip model agent task for pvc storage by @slin1237 in #171
- [core] add model metadata parser to ome agent by @slin1237 in #172
- [ci] update ci and makefile by @slin1237 in #173
- [utils] Add CoreWeave H200 shape configuration by @abatilo in #177
- Bump oraclelinux from 9-slim to 10-slim in /dockerfiles by @dependabot[bot] in #174
- fix missing the artifacts dir during
make artifacts
by @Iceber in #178
- [misc] fix incorrect integration path by @carlory in #179
- Remove training-agent references and update docs by @carlory in #180
- [core] add kimi k2 model support by @slin1237 in #182
- imporve supported runtime sort logic by @pallasathena92 in #183
- [misc] add kimi k2 runtime, isvc, and model by @slin1237 in #184
- [ci] add dev artifact publishing and fix openvex by @slin1237 in #187
- [core] add runtime selector package and deprecate runtime util by @slin1237 in #186
- [chore] update code owner by @slin1237 in #188
- fix: ๐ง Missing pkg/version package. by @yafengio in #190
- fix: Update status using Status().Update method by @carlory in #192
- fix: don't panic when webhook fails to get deploy config by @carlory in #193
- [ci] Fix dev chart handling in workflow by @carlory in #194
- [misc] fix external service being cleaned up and kimi runtime update by @slin1237 in #195
- [chore] update image for 0.1.3 release by @slin1237 in #196
๐ Changelog
Full Changelog: v0.1.2...v0.1.3
๐ Software Bill of Materials
SBOMs are available in both SPDX and CycloneDX formats for:
- Source code
- All container images
๐ Signatures
All container images are signed with cosign. Verify with:
cosign verify ghcr.io/moirai-internal/ome-manager:v0.1.3 --certificate-identity-regexp=https://github.com/sgl-project/ome/.github/workflows/release.yaml@refs/tags/.* --certificate-oidc-issuer=https://token.actions.githubusercontent.com
RetroSearch is an open source project built by @garambo
| Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4