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
2025-07-23 19:02:44 +03:00
parent 7a08e1a7f2
commit 444974a6ff
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sdnq/common.py
+1 -1
View File
@@ -31,7 +31,7 @@ if hasattr(torch, "float8_e4m3fnuz"):
if hasattr(torch, "float8_e5m2fnuz"):
dtype_dict["float8_e5m2fnuz"] = {"min": -57344, "max": 57344, "num_bits": 8, "target_dtype": "fp8", "torch_dtype": torch.float8_e5m2fnuz, "storage_dtype": torch.float8_e5m2fnuz, "is_unsigned": False, "is_integer": False}
use_tensorwise_fp8_matmul = True # Direct tensorwise only exist on H100 hardware, sdnq will use software implementation with this settig
use_tensorwise_fp8_matmul = True # Direct tensorwise only exist on H100 hardware, sdnq will use software tensorwise with this setting
quantized_matmul_dtypes = ("int8", "int7", "int6", "int5", "int4", "int3", "int2", "float8_e4m3fn", "float8_e5m2")
if devices.backend in {"cpu", "openvino"}:
quantized_matmul_dtypes += ("float8_e4m3fnuz", "float8_e5m2fnuz")