Skip to content

Commit 1e5bc3b

Browse files
authored
Move gemlite layout to prototype/dtypes (#3313)
1 parent 842a0c4 commit 1e5bc3b

File tree

7 files changed

+480
-449
lines changed

7 files changed

+480
-449
lines changed

torchao/dtypes/affine_quantized_tensor_ops.py

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -25,10 +25,6 @@
2525
_linear_f16_bf16_act_floatx_weight_check,
2626
_linear_f16_bf16_act_floatx_weight_impl,
2727
)
28-
from torchao.dtypes.uintx.gemlite_layout import (
29-
_linear_fp_act_int4_weight_gemlite_check,
30-
_linear_fp_act_int4_weight_gemlite_impl,
31-
)
3228
from torchao.dtypes.uintx.int4_cpu_layout import (
3329
_linear_fp_act_uint4_weight_cpu_check,
3430
_linear_fp_act_uint4_weight_cpu_impl,
@@ -90,6 +86,10 @@
9086
_linear_int8_act_int4_weight_cpu_check,
9187
_linear_int8_act_int4_weight_cpu_impl,
9288
)
89+
from torchao.prototype.dtypes.uintx.gemlite_layout import (
90+
_linear_fp_act_int4_weight_gemlite_check,
91+
_linear_fp_act_int4_weight_gemlite_impl,
92+
)
9393
from torchao.prototype.dtypes.uintx.marlin_qqq_tensor import (
9494
_linear_int8_act_int4_weight_marlin_qqq_check,
9595
_linear_int8_act_int4_weight_marlin_qqq_impl,

0 commit comments

Comments
 (0)