Merge branch 'nv_release_v2.8' into release_v2.8

transformer_engine/common/include/transformer_engine/recipe.h

@@ -124,6 +124,10 @@ void nvte_fp8_block_scaling_partial_cast(const NVTETensor inp, NVTETensor out,
                                          size_t start_offset, size_t block_len,
                                          const NVTEDType out_dtype, cudaStream_t stream);
 
+void nvte_nvfp4_compute_per_tensor_scale(const NVTETensor inpA, const bool use_rowwise_amax_A,
+                                         const NVTETensor inpB, const bool use_rowwise_amax_B,
+                                         float alpha_in, NVTETensor alpha_out, cudaStream_t stream);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif