Update rms_layernorm.py

This commit is contained in:
Daniel Han-Chen 2024-03-04 16:00:35 +11:00
commit 6fa081ae77

View file

@ -44,7 +44,7 @@ def _rms_layernorm_forward(
W_row = tl.load(W + col_offsets, mask = mask, other = 0)#.to(tl.float32)
row_var = tl.sum(X_row * X_row, axis = 0) / n_cols
inv_var = tl.math.rsqrt(row_var + eps)
inv_var = 1.0 / tl.math.sqrt(row_var + eps)
tl.store(r, inv_var)
normed = X_row * inv_var
normed = normed.to(W_row.dtype) # Exact copy from HF