Repository Issue Activity (beta)

#5128 CosineAnnealingLrScheduler displays non-standard behavior

#5127 Ability to chain LR schedulers to complement LR composition

#5124 LinearLrScheduler appears to produce wrong learning rates

#5122 `Fusion` produces wrong results for a batched MLP forward with a narrow `[N, 1]` output head (CUDA)

#4991 Explicit Stream Creation

#5091 Training Extension Causes Panic in UI code as a result of invalid progress rehydration

#4832 Add AdaptiveAvgPool3d module (adaptive 3D average pooling)

#5110 WebGPU wasm: Tensor::topk_with_indices panics with synchronous readback error

#4725 [Quantization] Fix sub-byte `QuantValue`s reshape

#1458 Implement prod and prod_dim for Autodiff backend

#5108 `any`/`all` reductions bypass the dedicated cubek reduce on fusion backends

#4793 tui panics when training

#5025 [cubecl] Device handle custom channel reduces all failures to opaque `CallError`

#5024 Autotune with fusion panics for output position out of bounds

#2406 Add `DataframeDataset` example

#4792 GPU memory leak during training

#5082 Training Continuation Causes Autodiff Error For Cuda and Pytorch Backends/Devices

#4836 Add a hypot function to float tensor ops

#5079 Vulkan stability problem while training with fusion

#5046 [feat] logical Any/All reduction (enables efficient bool_any / contains_nan)

#544 Training metrics

#5068 make ClassificationMetricConfig and associated structs public for custom configurations during training

#5069 Flex: broadcast element-wise ops miss all SIMD fast paths and fall back to a per-element scalar odometer

#5074 Pickle memo bomb — CPU + memory exhaustion (algorithmic-complexity DoS) in burn-store PyTorch reader

#4934 [Refactor] Backend generic removal