已合并
删除相关词汇 #1115
AtomGit-Bot创建于 2024年10月23日
删除相关词汇 #1115
已合并
AtomGit-Bot创建于 2024年10月23日
refs/pull/1115/head合入到master
3 个文件变更+3-5
Mmindspeed/core/tensor_parallel/tp_2d/linear_2d_split_along_first_dim.py+1-1
@@ -149,7 +149,7 @@ class Linear2DSplitAlongFirstDim(torch.autograd.Function):
149 use_bias = ctx.use_bias149 use_bias = ctx.use_bias
150 # first we prepare the total inputs needed to compute grad_input, grad_weight.150 # first we prepare the total inputs needed to compute grad_input, grad_weight.
151 # [s/(y*cp), b, E/x]---AG(y)---> [s/cp, b, E/x]151 # [s/(y*cp), b, E/x]---AG(y)---> [s/cp, b, E/x]
152- # Use sync AG to avoid communication competition, for the bandwidth is shared for 910C.152+ # Use sync AG to avoid communication competition, for the bandwidth is shared for A3.
153 grad_output = grad_output.contiguous()153 grad_output = grad_output.contiguous()
154 total_grad_output = sync_gather_along_first_dim(grad_output, ctx.rs_comm_intf)154 total_grad_output = sync_gather_along_first_dim(grad_output, ctx.rs_comm_intf)
155 155 
Mtests_extend/unit_tests/ops/cann/test_npu_all_to_all_all_gather_bmm.py+1-2
@@ -91,8 +91,7 @@ class TestNPUAlltoAllAllGatherBMM(DistributedTest):
91 hcomm_info_dist['tp_group'] = tp_group91 hcomm_info_dist['tp_group'] = tp_group
92 return hcomm_info_dist92 return hcomm_info_dist
93 93 
94- 94+ @pytest.mark.skipif(reason='device type is not supported, skip this UT!')
95- @pytest.mark.skipif(DEVICE_NAME != 'Ascend910C', reason='device type is not supported, skip this UT!')
96 @pytest.mark.parametrize('dtype', [torch.float16, torch.bfloat16])95 @pytest.mark.parametrize('dtype', [torch.float16, torch.bfloat16])
97 @pytest.mark.parametrize('out_y2_flag', [False, True])96 @pytest.mark.parametrize('out_y2_flag', [False, True])
98 @pytest.mark.parametrize('out_y3_flag', [False])97 @pytest.mark.parametrize('out_y3_flag', [False])
Mtests_extend/unit_tests/ops/cann/test_npu_bmm_reduce_scatter_all_to_all.py+1-2
@@ -82,8 +82,7 @@ class TestNPUBMMReduceScatterAlltoAll(DistributedTest):
82 hcomm_info_dist['tp_group'] = tp_group82 hcomm_info_dist['tp_group'] = tp_group
83 return hcomm_info_dist83 return hcomm_info_dist
84 84 
85- 85+ @pytest.mark.skipif(reason='device type is not supported, skip this UT!')
86- @pytest.mark.skipif(DEVICE_NAME != 'Ascend910C', reason='device type is not supported, skip this UT!')
87 @pytest.mark.parametrize('dtype', [torch.float16, torch.bfloat16])86 @pytest.mark.parametrize('dtype', [torch.float16, torch.bfloat16])
88 @pytest.mark.parametrize('y_shard_type', [1])87 @pytest.mark.parametrize('y_shard_type', [1])
89 @pytest.mark.parametrize('transpose_weight', [False, True])88 @pytest.mark.parametrize('transpose_weight', [False, True])