MQSMLA\QSMLA\SMLA lse bugfix-ops-transformer-AtomGit

已合并

MQSMLA\QSMLA\SMLA lse bugfix #9759

MQSMLA\QSMLA\SMLA lse bugfix #9759

已合并

Ccccccc创建于 19 天前

共 3 个文件变更+0-3

Mattention/mixed_quant_sparse_flash_mla/tests/pytest/mixed_quant_sparse_flash_mla_golden.py+0-1

	@@ -678,7 +678,6 @@ class GeneralizedSFAQuant:
678	b_index, n_index, :cur_act_seq, :	678	b_index, n_index, :cur_act_seq, :
679	]	679	]
680	t_start += cur_act_seq	680	t_start += cur_act_seq
681	- output = output.transpose(0, 1).contiguous()
682	return output	681	return output
683	else:	682	else:
684	return tensor	683	return tensor

Mattention/quant_sparse_flash_mla/tests/pytest/quant_sparse_flash_mla_golden.py+0-1

	@@ -594,7 +594,6 @@ class GeneralizedSFAQuant:
594	b_index, n_index, :cur_act_seq, :	594	b_index, n_index, :cur_act_seq, :
595	]	595	]
596	t_start += cur_act_seq	596	t_start += cur_act_seq
597	- output = output.transpose(0, 1).contiguous()
598	return output	597	return output
599	else:	598	else:
600	return tensor	599	return tensor

Mattention/sparse_flash_mla/tests/pytest/sparse_flash_mla_golden.py+0-1

	@@ -686,7 +686,6 @@ class GeneralizedSFA:
686	]	686	]
687	# output[t_start:t_end, n_index, :] = tensor[b_index, n_index, :cur_act_seq, :]	687	# output[t_start:t_end, n_index, :] = tensor[b_index, n_index, :cur_act_seq, :]
688	t_start += cur_act_seq	688	t_start += cur_act_seq
689	- output = output.transpose(0, 1).contiguous()
690	return output	689	return output
691	else:	690	else:
692	return tensor	691	return tensor