-
-
Notifications
You must be signed in to change notification settings - Fork 25
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[FEATURE] Decouple Foreman integrations from GitHub
enhancementNew feature or requestNew feature or requestStatus: Open.#1158 In defilantech/LLMKube;- Status: Open.#1157 In defilantech/LLMKube;
[FEATURE] Let ROCm address >64GB on Strix Halo (BIOS carveout + ttm.pages_limit retune) for large models
area/gpuGPU-related features and issuesGPU-related features and issuesenhancementNew feature or requestNew feature or requestgpuGPU acceleration featuresGPU acceleration featuresStatus: Open.#1155 In defilantech/LLMKube;Make the verify gate optional in the issue-batch decomposition (code → review)
enhancementNew feature or requestNew feature or requestStatus: Open.#1151 In defilantech/LLMKube;[FEATURE] SLO Grafana dashboard: support non-default windows (templated rule-name suffix)
area/observabilityMonitoring, metrics, logging, tracingMonitoring, metrics, logging, tracingenhancementNew feature or requestNew feature or requestStatus: Open.#1148 In defilantech/LLMKube;[FEATURE] Metal metrics federation so off-cluster InferenceServices back SLOs with data
area/observabilityMonitoring, metrics, logging, tracingMonitoring, metrics, logging, tracingenhancementNew feature or requestNew feature or requestStatus: Open.#1147 In defilantech/LLMKube;[FEATURE] request-latency histogram for the llama.cpp runtime (unblocks latency SLOs on llamacpp)
area/observabilityMonitoring, metrics, logging, tracingMonitoring, metrics, logging, tracingenhancementNew feature or requestNew feature or requestStatus: Open.#1146 In defilantech/LLMKube;[FEATURE] error_rate SLO indicator (blocked on per-request error metrics from serving runtimes)
area/observabilityMonitoring, metrics, logging, tracingMonitoring, metrics, logging, tracingenhancementNew feature or requestNew feature or requestStatus: Open.#1145 In defilantech/LLMKube;[BUG] InferenceService model-cache PVC claimName oscillates between two field managers → infinite Deployment rollout
bugSomething isn't workingSomething isn't workingStatus: Open.#1139 In defilantech/LLMKube;[BUG]
ModelRouteractivator serves a stale resident member and never self-heals after an out-of-band residency change (instant 502s until proxy restart)bugSomething isn't workingSomething isn't workingStatus: Open.#1123 In defilantech/LLMKube;[FEATURE] Edit-free nudge: count repeated read_file as no-progress so it can't be an escape hatch
area/foremanForeman: the agentic fleet orchestrator add-onForeman: the agentic fleet orchestrator add-onenhancementNew feature or requestNew feature or requestkind/featureNew feature or requestNew feature or requestpriority/mediumMedium priorityMedium prioritysize/smallSmall effort (< 1 day)Small effort (< 1 day)Status: Open.#1116 In defilantech/LLMKube;[FEATURE] ModelPool: cross-pod sticky model swapping on a shared GPU slot
area/gpuGPU-related features and issuesGPU-related features and issuesarea/routingMulti-backend routing, model router CRD, policy-aware dispatchMulti-backend routing, model router CRD, policy-aware dispatchenhancementNew feature or requestNew feature or requestkind/featureNew feature or requestNew feature or requestpriority/mediumMedium priorityMedium prioritysize/largeLarge effort (> 3 days)Large effort (> 3 days)Status: Open.