Add openpangu_moe model

docs/source/en/_toctree.yml

@@ -648,6 +648,8 @@
         title: OLMoE
       - local: model_doc/open-llama
         title: Open-Llama
+      - local: model_doc/openpangu_moe
+        title: OpenPanguMoE
       - local: model_doc/opt
         title: OPT
       - local: model_doc/pegasus

docs/source/en/model_doc/openpangu_moe.md

@@ -0,0 +1,71 @@
+<!--Copyright 2025 the HuggingFace Team. All rights reserved.
+
+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.
+
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer.
+
+-->
+
+
+# OpenPanguMoE
+
+## Overview
+
+The OpenPanguMoE model was proposed in [<INSERT PAPER NAME HERE>](<INSERT PAPER LINK HERE>) by <INSERT AUTHORS HERE>.
+<INSERT SHORT SUMMARY HERE>
+
+The abstract from the paper is the following:
+
+<INSERT PAPER ABSTRACT HERE>
+
+Tips:
+
+<INSERT TIPS ABOUT MODEL HERE>
+
+This model was contributed by [INSERT YOUR HF USERNAME HERE](https://huggingface.co/<INSERT YOUR HF USERNAME HERE>).
+The original code can be found [here](<INSERT LINK TO GITHUB REPO HERE>).
+
+## Usage examples
+
+<INSERT SOME NICE EXAMPLES HERE>
+
+## OpenpanguMoeConfig
+
+[[autodoc]] OpenpanguMoeConfig
+
+## OpenpanguMoeForCausalLM
+
+[[autodoc]] OpenpanguMoeForCausalLM
+
+## OpenpanguMoeModel
+
+[[autodoc]] OpenpanguMoeModel
+    - forward
+
+## OpenpanguMoePreTrainedModel
+
+[[autodoc]] OpenpanguMoePreTrainedModel
+    - forward
+
+## OpenpanguMoeForSequenceClassification
+
+[[autodoc]] OpenpanguMoeForSequenceClassification
+
+## OpenpanguMoeForQuestionAnswering
+
+[[autodoc]] OpenpanguMoeForQuestionAnswering
+
+## OpenpanguMoeForTokenClassification
+
+[[autodoc]] OpenpanguMoeForTokenClassification

src/transformers/models/__init__.py

@@ -256,6 +256,7 @@
     from .omdet_turbo import *
     from .oneformer import *
     from .openai import *
+    from .openpangu_moe import *
     from .opt import *
     from .ovis2 import *
     from .owlv2 import *

src/transformers/models/auto/configuration_auto.py

@@ -301,6 +301,7 @@
         ("oneformer", "OneFormerConfig"),
         ("open-llama", "OpenLlamaConfig"),
         ("openai-gpt", "OpenAIGPTConfig"),
+        ("openpangu_moe", "OpenPanguMoEConfig"),
         ("opt", "OPTConfig"),
         ("ovis2", "Ovis2Config"),
         ("owlv2", "Owlv2Config"),
@@ -763,6 +764,7 @@
         ("oneformer", "OneFormer"),
         ("open-llama", "OpenLlama"),
         ("openai-gpt", "OpenAI GPT"),
+        ("openpangu_moe", "OpenPanguMoE"),
         ("opt", "OPT"),
         ("ovis2", "Ovis2"),
         ("owlv2", "OWLv2"),

src/transformers/models/auto/modeling_auto.py

@@ -300,6 +300,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("oneformer", "OneFormerModel"),
         ("open-llama", "OpenLlamaModel"),
         ("openai-gpt", "OpenAIGPTModel"),
+        ("openpangu_moe", "OpenPanguMoEModel"),
         ("opt", "OPTModel"),
         ("ovis2", "Ovis2Model"),
         ("owlv2", "Owlv2Model"),
@@ -733,6 +734,7 @@ class _BaseModelWithGenerate(PreTrainedModel, GenerationMixin):
         ("olmoe", "OlmoeForCausalLM"),
         ("open-llama", "OpenLlamaForCausalLM"),
         ("openai-gpt", "OpenAIGPTLMHeadModel"),
+        ("openpangu_moe", "OpenPanguMoEForCausalLM"),
         ("opt", "OPTForCausalLM"),
         ("pegasus", "PegasusForCausalLM"),
         ("persimmon", "PersimmonForCausalLM"),

src/transformers/models/openpangu_moe/__init__.py

@@ -0,0 +1,29 @@
+# coding=utf-8
+# Copyright 2025 the HuggingFace Team. All rights reserved.
+#
+# 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 typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_openpangu_moe import *
+    from .modeling_openpangu_moe import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)

src/transformers/models/openpangu_moe/configuration_openpangu_moe.py

@@ -0,0 +1,81 @@
+# coding=utf-8
+# Copyright (c) 2025 Huawei Technologies Co., Ltd. All rights reserved.
+
+"""openPanguUltraMoE 718B model configuration"""
+
+from ...configuration_utils import PreTrainedConfig
+
+class OpenPanguMoEConfig(PreTrainedConfig):
+
+    model_type = "pangu_ultra_moe"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=153600,
+        hidden_size=7680,
+        intermediate_size=18432,
+        moe_intermediate_size=2048,
+        num_hidden_layers=61,
+        num_mtp_layers=1,
+        num_attention_heads=128,
+        num_key_value_heads=128,
+        num_shared_experts=1,
+        num_routed_experts=256,
+        routed_scaling_factor=2.5,
+        attention_kv_lora_dim=512,
+        attention_q_lora_dim=1536,
+        attention_qk_rope_dim=64,
+        attention_v_dim=128,
+        attention_qk_dim=128,
+        num_experts_per_tok=8,
+        num_dense_layers=3,
+        norm_topk_prob=True,
+        hidden_act="silu",
+        max_position_embeddings=131072,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=1,
+        tie_word_embeddings=False,
+        rope_theta=25600000,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+
+        self.num_dense_layers = num_dense_layers
+        self.intermediate_size = intermediate_size
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_shared_experts = num_shared_experts
+        self.num_routed_experts = num_routed_experts
+        self.routed_scaling_factor = routed_scaling_factor
+        self.num_experts_per_tok = num_experts_per_tok
+        self.norm_topk_prob = norm_topk_prob
+        self.attention_kv_lora_dim = attention_kv_lora_dim
+        self.attention_q_lora_dim = attention_q_lora_dim
+        self.attention_qk_rope_dim = attention_qk_rope_dim
+        self.attention_v_dim = attention_v_dim
+        self.attention_qk_dim = attention_qk_dim
+        self.attention_dropout = attention_dropout
+        self.num_mtp_layers = num_mtp_layers
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )