-
Notifications
You must be signed in to change notification settings - Fork 55
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
ci: stop waking the shared GB10 runner for changes that cannot affect what it builds
priority:mediumMedium priorityMedium prioritystatus:backlogIn the backlog, not yet readyIn the backlog, not yet readytype:choreMaintenance tasks (build, CI, etc.)Maintenance tasks (build, CI, etc.)Status: Open.#1992 In lablup/mlxcel;perf(models): maskless causal SDPA for prefill in Gemma 3, Gemma 4, Jamba, Nemotron-H
area:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadataplatform:macosmacOS (Apple Silicon) specificmacOS (Apple Silicon) specificpriority:mediumMedium priorityMedium prioritystatus:in-progressCurrently being worked onCurrently being worked ontype:performancePerformance improvementsPerformance improvementsStatus: Open.#1991 In lablup/mlxcel;fix: investigate Gemma 4 12B long-context MTP parity on Metal
area:inferenceGeneration, sampling, decoding (incl. speculative, DRY)Generation, sampling, decoding (incl. speculative, DRY)platform:macosmacOS (Apple Silicon) specificmacOS (Apple Silicon) specificpriority:highHigh priorityHigh prioritystatus:readyReady to be worked onReady to be worked ontype:bugBug fixes, error corrections, or issue resolutionsBug fixes, error corrections, or issue resolutionsStatus: Open.#1986 In lablup/mlxcel;fix: Reclaim paged blocks during decode when the pool budget is exhausted
priority:mediumMedium priorityMedium prioritystatus:in-progressCurrently being worked onCurrently being worked ontype:bugBug fixes, error corrections, or issue resolutionsBug fixes, error corrections, or issue resolutionsStatus: Open.#1982 In lablup/mlxcel;fix: Jamba v0.1 takes ~300 s for a 3.3k-token chat request on GB10
priority:mediumMedium priorityMedium prioritystatus:readyReady to be worked onReady to be worked ontype:performancePerformance improvementsPerformance improvementsStatus: Open.#1981 In lablup/mlxcel;fix(ci): clean up the Activity gate server and score on a quiet host
priority:mediumMedium priorityMedium prioritystatus:readyReady to be worked onReady to be worked ontype:bugBug fixes, error corrections, or issue resolutionsBug fixes, error corrections, or issue resolutionsStatus: Open.#1949 In lablup/mlxcel;chore(deps): bump the MLX pin and declare a base commit for each overlay
area:coremlxcel-core: MLX FFI, primitives, KV cache, layersmlxcel-core: MLX FFI, primitives, KV cache, layersplatform:linuxLinux (CUDA / packaging) specificLinux (CUDA / packaging) specificplatform:macosmacOS (Apple Silicon) specificmacOS (Apple Silicon) specificpriority:mediumMedium priorityMedium prioritystatus:readyReady to be worked onReady to be worked ontype:choreMaintenance tasks (build, CI, etc.)Maintenance tasks (build, CI, etc.)Status: Open.#1932 In lablup/mlxcel;ci(webui): the Activity performance gate fails on runner noise
priority:mediumMedium priorityMedium prioritystatus:readyReady to be worked onReady to be worked ontype:bugBug fixes, error corrections, or issue resolutionsBug fixes, error corrections, or issue resolutionsStatus: Open.#1925 In lablup/mlxcel;epic: make the bundled WebUI usable, on the shared ui-common foundation
area:architectureArchitecture and code structure changesArchitecture and code structure changespriority:highHigh priorityHigh prioritystatus:readyReady to be worked onReady to be worked ontype:enhancementNew features, capabilities, or significant additionsNew features, capabilities, or significant additionsStatus: Open.#1910 In lablup/mlxcel;docs(webui): ship release-time user docs and surface --webui in --help
area:cliCommand-line interface / CLI flagsCommand-line interface / CLI flagsarea:docsUser and developer documentationUser and developer documentationpriority:highHigh priorityHigh prioritystatus:readyReady to be worked onReady to be worked ontype:docsDocumentation improvements or additionsDocumentation improvements or additionsStatus: Open.#1900 In lablup/mlxcel;fix(rocm): the Gumbel-max and rejection sampler entry points have no ROCm arm and abort on metal_kernel
status:backlogIn the backlog, not yet readyIn the backlog, not yet readytype:bugBug fixes, error corrections, or issue resolutionsBug fixes, error corrections, or issue resolutionsStatus: Open.#1885 In lablup/mlxcel;fix(rocm): hipFFT blocks when too many plans are alive, root cause unknown
area:coremlxcel-core: MLX FFI, primitives, KV cache, layersmlxcel-core: MLX FFI, primitives, KV cache, layersplatform:linuxLinux (CUDA / packaging) specificLinux (CUDA / packaging) specificpriority:mediumMedium priorityMedium prioritystatus:readyReady to be worked onReady to be worked ontype:bugBug fixes, error corrections, or issue resolutionsBug fixes, error corrections, or issue resolutionsStatus: Open.#1876 In lablup/mlxcel;