-
Notifications
You must be signed in to change notification settings - Fork 52
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
Qwen3-VL vision tower is not token-exact against transformers on the image path
area:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadatamodeltype:vlmVision-language modelVision-language modelpriority: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.#1738 In lablup/mlxcel;feat(distributed): Kimi K3 pipeline stage executor and full-model multi-node validation
arch:hybridHybrid attention stack (linear/sliding/full or attention+SSM interleave)Hybrid attention stack (linear/sliding/full or attention+SSM interleave)area:inferenceGeneration, sampling, decoding (incl. speculative, DRY)Generation, sampling, decoding (incl. speculative, DRY)area:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadatamodelsize:xlarge4-bit checkpoint > 100GB; exceeds a 128GB dev box, needs bigger hardware4-bit checkpoint > 100GB; exceeds a 128GB dev box, needs bigger hardwaremodeltype:textText-only language modelText-only language modelpriority:lowLow priorityLow prioritystatus:blockedBlocked by dependencies or other issuesBlocked by dependencies or other issuestype:enhancementNew features, capabilities, or significant additionsNew features, capabilities, or significant additionsStatus: Open.#1734 In lablup/mlxcel;feat(models): add Arcee AFM-4.5B (arcee, dense ArceeForCausalLM) text model support
arch:denseDense transformer decoderDense transformer decoderarea:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadatamodelsize:small4-bit checkpoint <= 10GB; fast iteration, safe for smoke tests4-bit checkpoint <= 10GB; fast iteration, safe for smoke testsmodeltype:textText-only language modelText-only language modelpriority:mediumMedium priorityMedium prioritystatus:blockedBlocked by dependencies or other issuesBlocked by dependencies or other issuestype:enhancementNew features, capabilities, or significant additionsNew features, capabilities, or significant additionsStatus: Open.#1719 In lablup/mlxcel;perf(qwen_vl): multimodal RoPE leaks f32 into the decode residual
area:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadatamodeltype:vlmVision-language modelVision-language modelpriority:mediumMedium priorityMedium prioritystatus:reviewUnder reviewUnder reviewtype:performancePerformance improvementsPerformance improvementsStatus: Open.#1714 In lablup/mlxcel;perf(core): host graph build at decode is serial with device execution and scales with op count
arch:hybridHybrid attention stack (linear/sliding/full or attention+SSM interleave)Hybrid attention stack (linear/sliding/full or attention+SSM interleave)arch:moeSparse mixture-of-experts decoderSparse mixture-of-experts decoderarea:benchmarkBenchmark harness and performance measurement (bench_*.sh, /update-benchmarks)Benchmark harness and performance measurement (bench_*.sh, /update-benchmarks)area:coremlxcel-core: MLX FFI, primitives, KV cache, layersmlxcel-core: MLX FFI, primitives, KV cache, layersarea:inferenceGeneration, sampling, decoding (incl. speculative, DRY)Generation, sampling, decoding (incl. speculative, DRY)area:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadataplatform:macosmacOS (Apple Silicon) specificmacOS (Apple Silicon) specificpriority:highHigh priorityHigh prioritystatus:readyReady to be worked onReady to be worked ontype:performancePerformance improvementsPerformance improvementsStatus: Open.#1713 In lablup/mlxcel;test(turbo): delegated steel envelope fails intermittently with an identical RMS
area:coremlxcel-core: MLX FFI, primitives, KV cache, layersmlxcel-core: MLX FFI, primitives, KV cache, layerspriority:lowLow priorityLow prioritystatus:readyReady to be worked onReady to be worked ontype:testTest related changesTest related changesStatus: Open.#1712 In lablup/mlxcel;fix(phi,stablelm): port the upstream f32 attention guard
area:modelsModel architectures, weights, loading, metadataModel architectures, weights, loading, metadatapriority: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.#1710 In lablup/mlxcel;perf(core): bridge activation helpers promote half-precision to f32
area:benchmarkBenchmark harness and performance measurement (bench_*.sh, /update-benchmarks)Benchmark harness and performance measurement (bench_*.sh, /update-benchmarks)area:coremlxcel-core: MLX FFI, primitives, KV cache, layersmlxcel-core: MLX FFI, primitives, KV cache, layerspriority:highHigh priorityHigh prioritystatus:readyReady to be worked onReady to be worked ontype:performancePerformance improvementsPerformance improvementsStatus: Open.#1709 In lablup/mlxcel;docs(embeddings): the detection table omits model_type aliases the loader accepts
priority:lowLow priorityLow prioritystatus:readyReady to be worked onReady to be worked ontype:docsDocumentation improvements or additionsDocumentation improvements or additionsStatus: Open.#1700 In lablup/mlxcel;docs: CONTRIBUTING says clippy is not gated at PR time, but ci.yml runs -D warnings clippy on PRs
priority:mediumMedium priorityMedium prioritystatus:readyReady to be worked onReady to be worked ontype:docsDocumentation improvements or additionsDocumentation improvements or additionsStatus: Open.#1702 In lablup/mlxcel;docs: the code-guidelines HashMap-order section cites line numbers and counts that no longer exist
priority:lowLow priorityLow prioritystatus:readyReady to be worked onReady to be worked ontype:docsDocumentation improvements or additionsDocumentation improvements or additionsStatus: Open.#1706 In lablup/mlxcel;docs(install): the Cargo feature table omits three declared features
priority:lowLow priorityLow prioritystatus:readyReady to be worked onReady to be worked ontype:docsDocumentation improvements or additionsDocumentation improvements or additionsStatus: Open.#1705 In lablup/mlxcel;