Custom logsumexp (#2028)

* initial custom logsumexp

* more tests

* comments + fix
This commit is contained in:
Awni Hannun
2025-03-31 07:36:55 -07:00
committed by GitHub
parent ec2854b13a
commit de5f38fd48
27 changed files with 590 additions and 255 deletions

View File

@@ -40,7 +40,6 @@ template <typename T, typename AccT = T, int N_READS = SOFTMAX_N_READS>
local_max[simd_lane_id] = Limits<AccT>::min;
local_normalizer[simd_lane_id] = 0;
}
threadgroup_barrier(mem_flags::mem_threadgroup);
// Get the max
AccT maxval = Limits<AccT>::finite_min;