Search results

perf(cpu): SIMD and allocation optimizations for CPU compute path

Status: Open.

#221 In Samyssmile/edux;

· Samyssmile opened on Feb 28, 2026

feat(cuda): cache cuDNN descriptors for repeated operations

Status: Open.

#219 In Samyssmile/edux;

· Samyssmile opened on Feb 28, 2026

feat(cuda): enable TF32 Tensor Core math with graceful fallback for older GPUs

Status: Open.

#218 In Samyssmile/edux;

· Samyssmile opened on Feb 28, 2026

feat: Exponential Moving Average (EMA) of model weights

Status: Open.

#211 In Samyssmile/edux;

· Samyssmile opened on Feb 28, 2026

feat: Mixup and CutMix regularization for training

Status: Open.

#210 In Samyssmile/edux;

· Samyssmile opened on Feb 28, 2026

feat: Parameter groups — exclude BatchNorm from weight decay

Status: Open.

#209 In Samyssmile/edux;

· Samyssmile opened on Feb 28, 2026