-
Notifications
You must be signed in to change notification settings - Fork 1.3k
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[BUG]: LoRA adapter MDCs register the CLI KV block size, not the engine's main-attention block size — false 'all workers overloaded' 503s on hybrid-attention models
bugSomething isn't workingSomething isn't workingrouterRelates to routing, KV-aware routing, etc.Relates to routing, KV-aware routing, etc.Status: Open.#11866 In ai-dynamo/dynamo;DEP (light): Python-loadable frontend route extensions
enhancementNew feature or requestNew feature or requestfrontend`python -m dynamo.frontend` and `dynamo-run in=http|text|grpc``python -m dynamo.frontend` and `dynamo-run in=http|text|grpc`Status: Open.#11863 In ai-dynamo/dynamo;[BUG]: model advertised ready (/v1/models, has_any_ready_model) before generate endpoint is routable — requests 500 then 404 for seconds after startup
bugSomething isn't workingSomething isn't workingdynamo-llmRelates to dynamo-llm componentRelates to dynamo-llm componentdynamo-runtimeRelates to the dynamo-runtime componentRelates to the dynamo-runtime componentfrontend`python -m dynamo.frontend` and `dynamo-run in=http|text|grpc``python -m dynamo.frontend` and `dynamo-run in=http|text|grpc`Status: Open.#11855 In ai-dynamo/dynamo;feat(frontend): dedicated extension sub-router as the home for HTTP-level extension guardrails
enhancementNew feature or requestNew feature or requestfrontend`python -m dynamo.frontend` and `dynamo-run in=http|text|grpc``python -m dynamo.frontend` and `dynamo-run in=http|text|grpc`routerRelates to routing, KV-aware routing, etc.Relates to routing, KV-aware routing, etc.Status: Open.#11843 In ai-dynamo/dynamo;[FEATURE]: Add a single-GPU vLLM Snapshot + GMS recipe for GPT-OSS-120B
backend::vllmRelates to the vllm backendRelates to the vllm backenddynamo-runtimeRelates to the dynamo-runtime componentRelates to the dynamo-runtime componentStatus: Open.#11838 In ai-dynamo/dynamo;[BUG]: Multi-node DGD cannot be scaled to zero
backend::vllmRelates to the vllm backendRelates to the vllm backendbugSomething isn't workingSomething isn't workingdeployment::k8sRelates to dynamo deployment in kubernetesRelates to dynamo deployment in kubernetesdynamo-deployRelates to dynamo-deploy binaryRelates to dynamo-deploy binaryStatus: Open.#11837 In ai-dynamo/dynamo;[CONTRIBUTION]: Bound TCP accept failures under file-descriptor pressure
backend::vllmRelates to the vllm backendRelates to the vllm backendbugSomething isn't workingSomething isn't workingdynamo-runtimeRelates to the dynamo-runtime componentRelates to the dynamo-runtime componentfrontend`python -m dynamo.frontend` and `dynamo-run in=http|text|grpc``python -m dynamo.frontend` and `dynamo-run in=http|text|grpc`language::rustIssues/PRs that reference Rust codeIssues/PRs that reference Rust codeobservabilityRelated to metrics, tracing, loggingRelated to metrics, tracing, loggingperformanceInvolves the performance of DynamoInvolves the performance of DynamoStatus: Open.#11822 In ai-dynamo/dynamo;DevX improvement plan
enhancementNew feature or requestNew feature or requestStatus: Open.#11821 In ai-dynamo/dynamo;ai-dynamo frontend TCP accept-loop exhausts file descriptors and enters a runaway, unbounded retry loop under high concurrency, instead of backing off
backend::vllmRelates to the vllm backendRelates to the vllm backendbugSomething isn't workingSomething isn't workingdynamo-runtimeRelates to the dynamo-runtime componentRelates to the dynamo-runtime componentfrontend`python -m dynamo.frontend` and `dynamo-run in=http|text|grpc``python -m dynamo.frontend` and `dynamo-run in=http|text|grpc`language::rustIssues/PRs that reference Rust codeIssues/PRs that reference Rust codeobservabilityRelated to metrics, tracing, loggingRelated to metrics, tracing, loggingperformanceInvolves the performance of DynamoInvolves the performance of DynamoStatus: Open.#11801 In ai-dynamo/dynamo;[BUG]: ai-dynamo/vllm-runtime:1.3.0-dev.1-cuda13 ships a self-inconsistent nvidia_cutlass_dsl (missing
increment_coord) —import cutlassfailsbackend::vllmRelates to the vllm backendRelates to the vllm backendbugSomething isn't workingSomething isn't workingdynamo-runtimeRelates to the dynamo-runtime componentRelates to the dynamo-runtime componentStatus: Open.#11775 In ai-dynamo/dynamo;DEP (light): Experimental replicaGrouping for single-node Grove components
deployment::k8sRelates to dynamo deployment in kubernetesRelates to dynamo deployment in kubernetesdynamo-deployRelates to dynamo-deploy binaryRelates to dynamo-deploy binaryStatus: Open.#11774 In ai-dynamo/dynamo;feat(request-trace): native S3 sink for request-trace records
dynamo-llmRelates to dynamo-llm componentRelates to dynamo-llm componentlanguage::rustIssues/PRs that reference Rust codeIssues/PRs that reference Rust codeobservabilityRelated to metrics, tracing, loggingRelated to metrics, tracing, loggingStatus: Open.#11768 In ai-dynamo/dynamo;