feat(transformers/models): add models of eomt, timesfm

Author: The-truthh

mindone/transformers/__init__.py

@@ -483,6 +483,7 @@
 from .models.emu3 import Emu3ForCausalLM, Emu3ForConditionalGeneration, Emu3PreTrainedModel, Emu3TextModel, Emu3VQVAE
 from .models.encodec import EncodecModel, EncodecPreTrainedModel
 from .models.encoder_decoder import EncoderDecoderModel
+from .models.eomt import EomtForUniversalSegmentation, EomtImageProcessor, EomtImageProcessorFast, EomtPreTrainedModel
 from .models.ernie import (
     ErnieForCausalLM,
     ErnieForMaskedLM,
@@ -1331,6 +1332,7 @@
     TapasPreTrainedModel,
 )
 from .models.textnet import TextNetBackbone, TextNetForImageClassification, TextNetModel, TextNetPreTrainedModel
+from .models.timesfm import TimesFmModel, TimesFmModelForPrediction, TimesFmPreTrainedModel
 from .models.timesformer import TimesformerForVideoClassification, TimesformerModel, TimesformerPreTrainedModel
 from .models.trocr import TrOCRForCausalLM, TrOCRPreTrainedModel
 from .models.tvp import TvpForVideoGrounding, TvpModel, TvpPreTrainedModel

mindone/transformers/mindspore_adapter/utils.py

@@ -29,7 +29,10 @@
     "bool": ms.bool_,
 }
 
-
+_MIN_INT8 = ms.tensor(np.iinfo(np.int8).min, dtype=ms.int8)
+_MIN_INT16 = ms.tensor(np.iinfo(np.int16).min, dtype=ms.int16)
+_MIN_INT32 = ms.tensor(np.iinfo(np.int32).min, dtype=ms.int32)
+_MIN_INT64 = ms.tensor(np.iinfo(np.int64).min, dtype=ms.int64)
 _MIN_FP16 = ms.tensor(np.finfo(np.float16).min, dtype=ms.float16)
 _MIN_FP32 = ms.tensor(np.finfo(np.float32).min, dtype=ms.float32)
 _MIN_FP64 = ms.tensor(np.finfo(np.float64).min, dtype=ms.float64)
@@ -41,6 +44,10 @@
 
 
 _DTYPE_2_MIN = {
+    ms.int8: _MIN_INT8,
+    ms.int16: _MIN_INT16,
+    ms.int32: _MIN_INT32,
+    ms.int64: _MIN_INT64,
     ms.float16: _MIN_FP16,
     ms.float32: _MIN_FP32,
     ms.float64: _MIN_FP64,

mindone/transformers/models/__init__.py

@@ -72,6 +72,7 @@
     emu3,
     encodec,
     encoder_decoder,
+    eomt,
     ernie,
     esm,
     falcon,
@@ -224,6 +225,7 @@
     table_transformer,
     tapas,
     textnet,
+    timesfm,
     timesformer,
     trocr,
     tvp,

mindone/transformers/models/auto/configuration_auto.py

@@ -95,6 +95,7 @@
         ("emu3", "Emu3Config"),
         ("encodec", "EncodecConfig"),
         ("encoder-decoder", "EncoderDecoderConfig"),
+        ("eomt", "EomtConfig"),
         ("esm", "EsmConfig"),
         ("falcon", "FalconConfig"),
         ("falcon_mamba", "FalconMambaConfig"),
@@ -253,6 +254,7 @@
         ("table-transformer", "TableTransformerConfig"),
         ("tapas", "TapasConfig"),
         ("textnet", "TextNetConfig"),
+        ("timesfm", "TimesFmConfig"),
         ("timesformer", "TimesformerConfig"),
         ("trocr", "TrOCRConfig"),
         ("tvp", "TvpConfig"),
@@ -363,6 +365,7 @@
         ("emu3", "Emu3"),
         ("encodec", "Encodec"),
         ("encoder-decoder", "Encoder decoder"),
+        ("eomt", "EoMT"),
         ("esm", "ESM"),
         ("falcon", "Falcon"),
         ("falcon_mamba", "FalconMamba"),
@@ -525,6 +528,7 @@
         ("table-transformer", "Table Transformer"),
         ("tapas", "TAPAS"),
         ("textnet", "TextNet"),
+        ("timesfm", "TimesFm"),
         ("timesformer", "TimeSformer"),
         ("trocr", "TrOCR"),
         ("tvp", "TVP"),

mindone/transformers/models/auto/image_processing_auto.py

@@ -63,6 +63,7 @@
             ("dinov2", ("BitImageProcessor",)),
             ("dpt", ("DPTImageProcessor",)),
             ("efficientnet", ("EfficientNetImageProcessor",)),
+            ("eomt", ("EomtImageProcessor", "EomtImageProcessorFast")),
             ("flava", ("FlavaImageProcessor",)),
             ("llava_next", ("LlavaNextImageProcessor",)),
             ("llava_next_video", ("LlavaNextVideoImageProcessor",)),

mindone/transformers/models/auto/modeling_auto.py

@@ -230,6 +230,7 @@
         ("table-transformer", "TableTransformerModel"),
         ("tapas", "TapasModel"),
         ("textnet", "TextNetModel"),
+        ("timesfm", "TimesFmModel"),
         ("timesformer", "TimesformerModel"),
         ("tvp", "TvpModel"),
         ("udop", "UdopModel"),
@@ -653,6 +654,7 @@
     [
         # Model for Universal Segmentation mapping
         ("detr", "DetrForSegmentation"),
+        ("eomt", "EomtForUniversalSegmentation"),
         ("mask2former", "Mask2FormerForUniversalSegmentation"),
         ("maskformer", "MaskFormerForInstanceSegmentation"),
         ("oneformer", "OneFormerForUniversalSegmentation"),
@@ -1268,6 +1270,12 @@
 
 MODEL_FOR_TIME_SERIES_REGRESSION_MAPPING_NAMES = OrderedDict()
 
+MODEL_FOR_TIME_SERIES_PREDICTION_MAPPING_NAMES = OrderedDict(
+    [
+        ("timesfm", "TimesFmModelForPrediction"),
+    ]
+)
+
 MODEL_FOR_IMAGE_TO_IMAGE_MAPPING_NAMES = OrderedDict(
     [
         ("swin2sr", "Swin2SRForImageSuperResolution"),

mindone/transformers/models/eomt/__init__.py

@@ -0,0 +1,19 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# This code is adapted from https://github.com/huggingface/transformers
+# with modifications to run transformers on mindspore.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .image_processing_eomt import *
+from .image_processing_eomt_fast import *
+from .modeling_eomt import *