Repository Issue Activity (beta)

#2287 Switch ModelMirror to be namespace-scoped

#2258 Surface model mirror and streaming errors and metrics in status

#2286 feat: preset-driven speculative decoding toggle for Workspace inference

#2285 feat: support preferred instanceType in BYO mode

#2282 nvme disk missing when creating node with karpenter

#2276 chore: url check

#2268 chore: placeholder

#2191 get_max_gpu_memory_utilization() conflicts with a hardcoded --gpu-memory-utilization=0.84

#2257 Report actual max-model-len in Workspace status

#2253 Expose terminationGracePeriodSeconds in InferenceSetTemplate so users can set a grace period appropriate for their model and workload:

#2252 Make Flux optional for GWIE chart deployment

#2114 Azure GPU SKU mapping mismatch and vLLM runtime failure on Standard_NC6s_v3 / Standard_NC4as_T4_v3

#2220 support custom node taints beyond spot instances

#2202 [Feature] Improve sas fetching logic

#2218 Configurable OCI source and image registry for InferenceSet auto-provisioned InferencePool/GW extension

#2181 InferenceSet CRD requires instanceType but Workspace webhook rejects it in BYO mode (nodeProvisioner=byo)

#1744 Feature Request: Support for NVIDIA Multi-Instance GPU (MIG)

#2194 feat: support workload-identity-authed storage class for model mirror and streaming

#2196 Improve drift-driven node image upgrade for Karpenter node provisioner

#2165 Stream model from url in annotation

#2154 [VOTE NEEDED]: Moving AI Runway repository to a new org requires formal vote from Kaito maintainers

#2174 BYO mode: InferenceSet multi-node workspaces can deadlock when competing for the same nodepool

#1295 Support disaggregated serving

#2074 Expose model download progress metrics and surface remaining download time in Workspace / InferenceSet status

#2057 Model mirror and streaming