Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 01:04:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

cleanup

Browse Source
This commit is contained in:
Disty0
2026-06-23 20:44:37 +03:00
parent 0af57a0af1
commit e7f11ed772
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sdnq/kernels/triton_atten.py
+1 -2
View File
@@ -1,4 +1,3 @@
import os
import math
import torch
@@ -64,7 +63,7 @@ def quantize_attn(
q_q, q_scale = quantize_tensor(q, group_size=group_size, matmul_dtype=matmul_dtype)
k_q, k_scale = quantize_tensor(k, group_size=group_size_kv, matmul_dtype=matmul_dtype)
v_q, v_scale = quantize_tensor(v, group_size=group_size_kv, matmul_dtype=pv_matmul_dtype)
q_scale = q_scale.mul_(q.shape[-1]**-0.5 * 1.4426950408889634)
q_scale = q_scale.mul_(scale * 1.4426950408889634)
return q_q, q_scale, k_q, k_scale, v_q, v_scale