GitLab

* not fuse bias for output all reduction case + unit tests
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* norm to reduce dgamma along tpsp as well
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* clean up tests
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* fix test_distributed_layernorm byte counts
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

* increase tols for jax_gemm
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>

---------
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>