GitLab

* feat: add triton kernels to decrease latency of large batches

* cast to int32

* fix kernel

* fix kernel

* disable triton on rocm

* fix speculation

* add slots filtering kernel