Remove triton per group quant [ROCm] [Bugfix] (#49621)
Signed-off-by: Andy Friedrich <afriedri@amd.com>
This commit is contained in:
@@ -195,8 +195,6 @@ class TestModel(torch.nn.Module):
|
||||
# Blockwise path
|
||||
if self.use_aiter_fusion and self.use_aiter_quant_op:
|
||||
return [rocm_aiter_ops.get_group_quant_op()]
|
||||
if self.use_aiter_fusion:
|
||||
return [torch.ops.vllm.triton_per_token_group_quant_fp8.default]
|
||||
else:
|
||||
if self.use_aiter_quant_op:
|
||||
return [rocm_aiter_ops.get_per_token_quant_op()]
|
||||
|
||||
Reference in New Issue
Block a user