Remove hardcoded device="cuda" to support more devices (#2503)

Co-authored-by: Jiang Li <jiang1.li@intel.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
Kunshang Ji
2024-02-01 15:46:39 -08:00
committed by GitHub
co-authored by Jiang Li
parent c410f5d020
commit 96b6f475dd
32 changed files with 343 additions and 292 deletions
@@ -127,7 +127,6 @@ class GPTQLinearMethod(LinearMethodBase):
torch.empty(
input_size_per_partition // self.quant_config.pack_factor,
output_size_per_partition,
device="cuda",
dtype=torch.int32,
),
requires_grad=False,
@@ -145,7 +144,6 @@ class GPTQLinearMethod(LinearMethodBase):
i // self.quant_config.group_size
for i in range(input_size_per_partition)
],
device="cuda",
dtype=torch.int32,
),
requires_grad=False,
@@ -156,7 +154,6 @@ class GPTQLinearMethod(LinearMethodBase):
torch.empty(
scale_and_zero_size,
output_size_per_partition // self.quant_config.pack_factor,
device="cuda",
dtype=torch.int32,
),
requires_grad=False,
@@ -172,7 +169,6 @@ class GPTQLinearMethod(LinearMethodBase):
torch.empty(
scale_and_zero_size,
output_size_per_partition,
device="cuda",
dtype=params_dtype,
),
requires_grad=False,