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

Enable Triton MM for Intel

Browse Source
This commit is contained in:
Disty0
2026-07-13 18:24:14 +03:00
parent 32650c7e50
commit 78482ef6ca
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sdnq/kernel_wrappers.py
+5 -5
View File
@@ -38,7 +38,7 @@ else:
use_openvino_mm = bool(os.environ.get("SDNQ_USE_OPENVINO_MM", "0").lower() not in {"0", "false", "no"})
if os.environ.get("SDNQ_USE_TRITON_MM", None) is None:
use_triton_mm = bool(is_rdna2_and_older or devices.backend == "zluda")
use_triton_mm = bool(is_rdna2_and_older or devices.backend in {"zluda", "ipex"})
else:
use_triton_mm = bool(os.environ.get("SDNQ_USE_TRITON_MM", "0").lower() not in {"0", "false", "no"})
@@ -68,7 +68,6 @@ if use_openvino_mm:
from .kernels.openvino_mm import openvino_int_mm, openvino_fp_mm
int_mm_func = openvino_int_mm
fp_mm_func = openvino_fp_mm
fp8_mm_func = openvino_fp_mm
except Exception:
use_openvino_mm = False
elif use_triton_mm:
@@ -77,11 +76,12 @@ elif use_triton_mm:
from .kernels.triton_scaled_mm import sdnq_scaled_mm
int_mm_func = sdnq_triton_mm
fp_mm_func = sdnq_triton_mm
fp8_mm_func = sdnq_triton_mm
int_scaled_mm_func = sdnq_scaled_mm
fp_scaled_mm_func = sdnq_scaled_mm
fp8_scaled_mm_func = sdnq_scaled_mm
use_tensorwise_fp8_matmul = False
if is_fp8_mm_supported:
fp8_mm_func = sdnq_triton_mm
fp8_scaled_mm_func = sdnq_scaled_mm
use_tensorwise_fp8_matmul = False
except Exception:
use_triton_mm = False