Merge branch 'mindspore-lab:master' into master

Author: linrx-ctrl

mindnlp/peft_lora_mindnlp.ipynb

@@ -152,6 +152,7 @@
     mobilenet_v1,
     mobilenet_v2,
     mobilevit,
+    mobilevitv2,
     mpnet,
     mpt,
     mllama,
@@ -395,6 +396,7 @@
 from .mobilenet_v1 import *
 from .mobilenet_v2 import *
 from .mobilevit import *
+from .mobilevitv2 import *
 from .mpnet import *
 from .mllama import *
 from .mluke import *
@@ -640,6 +642,7 @@
 __all__.extend(mobilenet_v1.__all__)
 __all__.extend(mobilenet_v2.__all__)
 __all__.extend(mobilevit.__all__)
+__all__.extend(mobilevitv2.__all__)
 __all__.extend(mpnet.__all__)
 __all__.extend(mpt.__all__)
 __all__.extend(mt5.__all__)

mindnlp/peft_lora_pytorch.ipynb

@@ -468,6 +468,7 @@
         ("ctrl", "CTRLLMHeadModel"),
         ("data2vec-text", "Data2VecTextForCausalLM"),
         ("dbrx", "DbrxForCausalLM"),
+        ("deepseek_v2", "DeepseekV2ForCausalLM"),
         ("electra", "ElectraForCausalLM"),
         ("ernie", "ErnieForCausalLM"),
         ("falcon", "FalconForCausalLM"),

mindnlp/transformers/models/__init__.py

@@ -470,13 +470,13 @@ def bigbird_block_sparse_attention(
             [
                 to_mask[:, :, :, : 3 * to_block_size],
                 to_mask[:, :, :, -to_block_size:],
-                ops.ones([bsz, 1, 1, n_rand_blocks * to_block_size], dtype=to_mask.dtype),
+                ops.ones((bsz, 1, 1, n_rand_blocks * to_block_size), dtype=to_mask.dtype),
             ],
             dim=3,
         )
         second_rand_pad = ops.cat(
             [
-                ops.ones([bsz, n_heads, from_block_size, 4 * to_block_size], dtype=rand_mask.dtype),
+                ops.ones((bsz, n_heads, from_block_size, 4 * to_block_size), dtype=rand_mask.dtype),
                 rand_mask[:, :, 0],
             ],
             dim=3,
@@ -604,13 +604,13 @@ def bigbird_block_sparse_attention(
             [
                 to_mask[:, :, :, :to_block_size],
                 to_mask[:, :, :, -3 * to_block_size :],
-                ops.ones([bsz, 1, 1, n_rand_blocks * to_block_size], dtype=to_mask.dtype),
+                ops.ones((bsz, 1, 1, n_rand_blocks * to_block_size), dtype=to_mask.dtype),
             ],
             dim=3,
         )
         second_last_rand_pad = ops.cat(
             [
-                ops.ones([bsz, n_heads, from_block_size, 4 * to_block_size], dtype=rand_mask.dtype),
+                ops.ones((bsz, n_heads, from_block_size, 4 * to_block_size), dtype=rand_mask.dtype),
                 rand_mask[:, :, -1],
             ],
             dim=3,

mindnlp/transformers/models/auto/modeling_auto.py

@@ -450,13 +450,13 @@ def bigbird_block_sparse_attention(
             [
                 to_mask[:, :, :, : 3 * to_block_size],
                 to_mask[:, :, :, -to_block_size:],
-                ops.ones([bsz, 1, 1, n_rand_blocks * to_block_size], dtype=to_mask.dtype),
+                ops.ones((bsz, 1, 1, n_rand_blocks * to_block_size), dtype=to_mask.dtype),
             ],
             dim=3,
         )
         second_rand_pad = ops.cat(
             [
-                ops.ones([bsz, n_heads, from_block_size, 4 * to_block_size], dtype=rand_mask.dtype),
+                ops.ones((bsz, n_heads, from_block_size, 4 * to_block_size), dtype=rand_mask.dtype),
                 rand_mask[:, :, 0],
             ],
             dim=3,
@@ -584,13 +584,13 @@ def bigbird_block_sparse_attention(
             [
                 to_mask[:, :, :, :to_block_size],
                 to_mask[:, :, :, -3 * to_block_size :],
-                ops.ones([bsz, 1, 1, n_rand_blocks * to_block_size], dtype=to_mask.dtype),
+                ops.ones((bsz, 1, 1, n_rand_blocks * to_block_size), dtype=to_mask.dtype),
             ],
             dim=3,
         )
         second_last_rand_pad = ops.cat(
             [
-                ops.ones([bsz, n_heads, from_block_size, 4 * to_block_size], dtype=rand_mask.dtype),
+                ops.ones((bsz, n_heads, from_block_size, 4 * to_block_size), dtype=rand_mask.dtype),
                 rand_mask[:, :, -1],
             ],
             dim=3,