L-electron-Rare/mlx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[CUDA] Fallback QMM (#3315)

Browse Source
This commit is contained in:
Cheng
2026-04-01 12:41:26 +09:00
committed by GitHub
parent 7cd73c4202
commit 1c9ee2f655
16 changed files with 700 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/tests/cuda_skip.py
-6
View File
@@ -1,5 +1,4 @@
cuda_skip = {
"TestLayers.test_quantized_embedding",
# Gather matmul NYI
"TestBlas.test_gather_matmul",
"TestBlas.test_gather_matmul_grad",
@@ -25,16 +24,11 @@ cuda_skip = {
"TestQuantized.test_gather_qmm_sorted",
"TestQuantized.test_gather_qmm_grad",
"TestQuantized.test_non_multiples",
"TestQuantized.test_qmm",
"TestQuantized.test_qmm_jvp",
"TestQuantized.test_qmm_shapes",
"TestQuantized.test_qmm_vjp",
"TestQuantized.test_fp_qvm",
"TestQuantized.test_qvm",
"TestQuantized.test_qvm_splitk",
"TestQuantized.test_qmv_small_non_multiples",
"TestQuantized.test_small_matrix",
"TestQuantized.test_throw",
"TestQuantized.test_vjp_scales_biases",
"TestExportImport.test_export_quantized_model",
}