Runtime check for the FlashAttention version rejects local version identifiers when it shouldn't

[BUG] NVFP4BlockScaling forward pass fails on RTX 5080 (sm_120) when K is a multiple of 128; sm_100/B100 TMEM+UMMA hardcoded in row_cast_col_hadamard_transform_cast_fusion.cu

[BUG]TensorRT engine export fails with UNSUPPORTED_NODE errors after TE FP8 training

[Bug] moe_permute CUDA kernel: int32 overflow and incorrect -1 sentinel handling

FP8 weight caching shows no speedup (sometimes slowdown) and causes training curve differences under Float8BlockScaling

Request for batched general_gemm() (or FP8-aware torch.bmm) for non-Linear GEMM workloads

[Core] MXFP8 grouped quantization kernel crashes with groups of size zero

undefined symbol: _ZN3c104cuda29c10_cuda_check_implementationEiPKcS2_ib

Grouped Bias/Dbias Kernel Support After Grouped GEMM

[Question] Expected behavior for blockwise FP8? Hybrid E4M3/E5M2 format & eval metrics outperforming BF16