Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-20 01:31:13 +02:00
Code Issues Packages Projects Releases Wiki Activity

Don't downcast scale with fp8 matmul

Browse Source
This commit is contained in:
Disty0
2025-05-29 16:35:40 +03:00
parent 2351efb8f7
commit df8b31fcfc
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant_sdnq.py
+1 -1
View File
@@ -117,7 +117,7 @@ def sdnq_quantize_layer(layer, weights_dtype="int8", torch_dtype=None, group_siz
zero_point = None
layer.weight.data = quantize_weight(layer.weight, scale, zero_point, weights_dtype)
if not shared.opts.sdnq_decompress_fp32:
if not shared.opts.sdnq_decompress_fp32 and not (use_quantized_matmul and not dtype_dict[weights_dtype]["is_integer"]):
scale = scale.to(torch_dtype)
if zero_point is not None:
zero_point = zero_point.to(torch_dtype)