burn · Issues· 311 open

Question: burn-router vs burn-dispatch

Feature request - allow reuse of Gradients

Feature Request - lifecycle callbacks for SupervisedTraining

Gradient accumulation drops final partial window and advances LR scheduler per microbatch

burn-store: Unsound clone_unsafely visitor memcpy causes double-free/UB, and deserializer panics on unsupported types

[burn 0.21 / burn-fusion] Feature request: public API to construct a FusionTensor from a base-backend tensor (Client::adopt_base_tensor)

burn-ndarray: relu and clamp drop NaN once a tensor is long enough to vectorize

pytorch-reader: LegacySource reopens the checkpoint by path on every read

pytorch-reader: checkpoints saved with compute_crc32=False fail to open with "Invalid checksum"

pytorch-reader: README lists three containers, lib.rs documents four

pytorch-reader: read a checkpoint from bytes, not only from a path

pytorch-reader: PytorchError has no source() chain, doubled hedges in Display, and a non-checkpoint file reports a raw opcode

pytorch-reader: ZIP tensor reads serialize on a Mutex around the archive; 8 threads read no faster than 1

burn-optim: FP16 norm clipping can zero finite gradients; FP32 promotion or max scaling?

Add automatic mixed precision training with BF16 autocast and FP16 gradient scaling