-
Notifications
You must be signed in to change notification settings - Fork 15
All issues
Issue creation is restricted in this repository
- #1237 · michalharakal opened
on Aug 31, 2026
Issues
is:issue state:open
is:issue state:open
Search results
Tracking: Whisper-class offline ASR on Android — engine-side P0
compute-backendCompute backends and graphsCompute backends and graphsskill:androidAndroid target/build/kernel workAndroid target/build/kernel worktrackingParent/tracking issue with sub-issuesParent/tracking issue with sub-issuesStatus: Open.#1282 In SKaiNET-developers/SKaiNET;FP16 simulation execution context + FP32-accumulating LayerNorm lowering rule (variance overflow repro)
enhancementNew feature or requestNew feature or requestskill:numericsPyTorch/NumPy/math background, no Kotlin requiredPyTorch/NumPy/math background, no Kotlin requiredStatus: Open.#1281 In SKaiNET-developers/SKaiNET;TensorDump forward hook + TensorCompare CLI: localise numeric divergence between two runs
enhancementNew feature or requestNew feature or requestskill:kotlin-coreKotlin implementation in commonMainKotlin implementation in commonMainskill:numericsPyTorch/NumPy/math background, no Kotlin requiredPyTorch/NumPy/math background, no Kotlin requiredStatus: Open.#1280 In SKaiNET-developers/SKaiNET;ONNX importer: LSTM, Gather, Unsqueeze/Squeeze, Cast, Where, If (Silero VAD import as the gate)
enhancementNew feature or requestNew feature or requestfile-format:onnxONNX file formatONNX file formatStatus: Open.#1279 In SKaiNET-developers/SKaiNET;Native Silero VAD model on Lstm with weights from the upstream ONNX file
enhancementNew feature or requestNew feature or requestmodelGeneral model supportGeneral model supportskill:numericsPyTorch/NumPy/math background, no Kotlin requiredPyTorch/NumPy/math background, no Kotlin requiredStatus: Open.#1278 In SKaiNET-developers/SKaiNET;Schedule on Android: verify CoroutineSchedule on ART and measure scheduled SDPA on device
compute-backendCompute backends and graphsCompute backends and graphsskill:androidAndroid target/build/kernel workAndroid target/build/kernel workStatus: Open.#1277 In SKaiNET-developers/SKaiNET;Native fused LayerNorm / GELU / softmax / conv1d kernels for the JNI and FFM tiers
compute-backendCompute backends and graphsCompute backends and graphsenhancementNew feature or requestNew feature or requestskill:androidAndroid target/build/kernel workAndroid target/build/kernel workskill:nativeKotlin/Native (Linux, macOS) or FFM kernel workKotlin/Native (Linux, macOS) or FFM kernel workStatus: Open.#1276 In SKaiNET-developers/SKaiNET;Dense FP32 (and FP16) matmul on the Android JNI tier: attention runs scalar today
compute-backendCompute backends and graphsCompute backends and graphsenhancementNew feature or requestNew feature or requestskill:androidAndroid target/build/kernel workAndroid target/build/kernel workskill:nativeKotlin/Native (Linux, macOS) or FFM kernel workKotlin/Native (Linux, macOS) or FFM kernel workStatus: Open.#1275 In SKaiNET-developers/SKaiNET;Panama FP32 matmul: first call in a JVM differs by an ULP from later calls (JIT-dependent reduceLanes order)
bugSomething isn't workingSomething isn't workingcompute-backendCompute backends and graphsCompute backends and graphsskill:numericsPyTorch/NumPy/math background, no Kotlin requiredPyTorch/NumPy/math background, no Kotlin requiredStatus: Open.#1261 In SKaiNET-developers/SKaiNET;👋 New contributors: start here
documentationImprovements or additions to documentation (DARC: D)Improvements or additions to documentation (DARC: D)help wantedExtra attention is neededExtra attention is neededStatus: Open.#1237 In SKaiNET-developers/SKaiNET;[Lane 6 · review] Precision/Recall/F1 — DARC review and @DarcValidated
assessmentAssessment task (DARC: A)Assessment task (DARC: A)size:sA few hoursA few hoursskill:reviewDARC review; must not be the task's implementerDARC review; must not be the task's implementersub-issueSub-issue of a tracking issueSub-issue of a tracking issueStatus: Open.#1235 In SKaiNET-developers/SKaiNET;[Lane 5 · docs] Precision/Recall/F1 — doc partials (math / intuition / examples / references)
documentationImprovements or additions to documentation (DARC: D)Improvements or additions to documentation (DARC: D)good first issueGood for newcomers; no prior SKaiNET codebase knowledge assumedGood for newcomers; no prior SKaiNET codebase knowledge assumedsize:sA few hoursA few hoursskill:docsAsciiDoc / technical writingAsciiDoc / technical writingsub-issueSub-issue of a tracking issueSub-issue of a tracking issueStatus: Open.#1234 In SKaiNET-developers/SKaiNET;