diff --git a/.gitattributes b/.gitattributes
index a6344aac8c09253b3b630fb776ae94478aa0275b..52373fe24473b1aa44333d318f578ae6bf04b49b 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text
diff --git a/chat_template.jinja b/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..41478957aca7a04b7321022e7d1f73de5badd995
--- /dev/null
+++ b/chat_template.jinja
@@ -0,0 +1,103 @@
+[gMASK]<sop>
+{%- if tools -%}
+<|system|>
+# Tools
+
+You may call one or more functions to assist with the user query.
+
+You are provided with function signatures within <tools></tools> XML tags:
+<tools>
+{% for tool in tools %}
+{{ tool | tojson(ensure_ascii=False) }}
+{% endfor %}
+</tools>
+
+For each function call, output the function name and arguments within the following XML format:
+<tool_call>{function-name}
+<arg_key>{arg-key-1}</arg_key>
+<arg_value>{arg-value-1}</arg_value>
+<arg_key>{arg-key-2}</arg_key>
+<arg_value>{arg-value-2}</arg_value>
+...
+</tool_call>{%- endif -%}
+{%- macro visible_text(content) -%}
+    {%- if content is string -%}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping -%}
+        {%- for item in content -%}
+            {%- if item is mapping and item.type == 'text' -%}
+                {{- item.text }}
+            {%- elif item is string -%}
+                {{- item }}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{- content }}
+    {%- endif -%}
+{%- endmacro -%}
+{%- set ns = namespace(last_user_index=-1) %}
+{%- for m in messages %}
+    {%- if m.role == 'user' %}
+        {% set ns.last_user_index = loop.index0 -%}
+    {%- endif %}
+{%- endfor %}
+{% for m in messages %}
+{%- if m.role == 'user' -%}<|user|>
+{{ visible_text(m.content) }}
+{{- '/nothink' if (enable_thinking is defined and not enable_thinking and not visible_text(m.content).endswith("/nothink")) else '' -}}
+{%- elif m.role == 'assistant' -%}
+<|assistant|>
+{%- set reasoning_content = '' %}
+{%- set content = visible_text(m.content) %}
+{%- if m.reasoning_content is string %}
+    {%- set reasoning_content = m.reasoning_content %}
+{%- else %}
+    {%- if '</think>' in content %}
+        {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+        {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+    {%- endif %}
+{%- endif %}
+{%- if loop.index0 > ns.last_user_index and reasoning_content -%}
+{{ '\n<think>' + reasoning_content.strip() +  '</think>'}}
+{%- else -%}
+{{ '\n<think></think>' }}
+{%- endif -%}
+{%- if content.strip() -%}
+{{ '\n' + content.strip() }}
+{%- endif -%}
+{% if m.tool_calls %}
+{% for tc in m.tool_calls %}
+{%- if tc.function %}
+    {%- set tc = tc.function %}
+{%- endif %}
+{{ '\n<tool_call>' + tc.name }}
+{% set _args = tc.arguments %}
+{% for k, v in _args.items() %}
+<arg_key>{{ k }}</arg_key>
+<arg_value>{{ v | tojson(ensure_ascii=False) if v is not string else v }}</arg_value>
+{% endfor %}
+</tool_call>{% endfor %}
+{% endif %}
+{%- elif m.role == 'tool' -%}
+{%- if m.content is string -%}
+{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+    {{- '<|observation|>' }}
+{%- endif %}
+{{- '\n<tool_response>\n' }}
+{{- m.content }}
+{{- '\n</tool_response>' }}
+{%- else -%}
+<|observation|>{% for tr in m.content %}
+
+<tool_response>
+{{ tr.output if tr.output is defined else tr }}
+</tool_response>{% endfor -%}
+{% endif -%}
+{%- elif m.role == 'system' -%}
+<|system|>
+{{ visible_text(m.content) }}
+{%- endif -%}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    <|assistant|>{{- '\n<think></think>' if (enable_thinking is defined and not enable_thinking) else '' -}}
+{%- endif -%}
\ No newline at end of file
diff --git a/config.json b/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..95ce6f89fbd9ccaf18cf2c37659c4d7e61e2c643
--- /dev/null
+++ b/config.json
@@ -0,0 +1,48 @@
+{
+  "architectures": [
+    "Glm4SharedMoeForCausalLM"
+  ],
+  "attention_bias": true,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_glm4_shared_moe.Glm4SharedMoeConfig",
+    "AutoModel": "modeling_glm4_shared_moe.Glm4SharedMoeModel",
+    "AutoModelForCausalLM": "modeling_glm4_shared_moe.Glm4SharedMoeForCausalLM"
+  },
+  "eos_token_id": [
+    151329,
+    151336,
+    151338
+  ],
+  "first_k_dense_replace": 1,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 10944,
+  "max_position_embeddings": 131072,
+  "model_type": "glm4_shared_moe",
+  "moe_intermediate_size": 1408,
+  "n_group": 1,
+  "n_routed_experts": 128,
+  "n_shared_experts": 1,
+  "norm_topk_prob": true,
+  "num_attention_heads": 96,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 46,
+  "num_key_value_heads": 8,
+  "num_nextn_predict_layers": 1,
+  "pad_token_id": 151329,
+  "partial_rotary_factor": 0.5,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "routed_scaling_factor": 1.0,
+  "tie_word_embeddings": false,
+  "topk_group": 1,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.54.1",
+  "use_cache": true,
+  "use_qk_norm": false,
+  "vocab_size": 151552
+}
diff --git a/configuration_glm4_shared_moe.py b/configuration_glm4_shared_moe.py
new file mode 100644
index 0000000000000000000000000000000000000000..c273c214275486b5e9cbf02c99e771f8642e679a
--- /dev/null
+++ b/configuration_glm4_shared_moe.py
@@ -0,0 +1,236 @@
+# coding=utf-8
+# Copyright 2025 The ZhipuAI Inc. team and HuggingFace Inc. 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 transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+
+
+class Glm4SharedMoeConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`Glm4MoeModel`]. It is used to instantiate a
+    Glm4Moe model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of [THUDM/GLM-4-100B-A10B](https://huggingface.co/THUDM/GLM-4-100B-A10B).
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 151552):
+            Vocabulary size of the Glm4Moe model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`Glm4MoeModel`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 10944):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 46):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 96):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        partial_rotary_factor (`float`, *optional*, defaults to 0.5):
+            The factor of the partial rotary position.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details, check out [this
+            paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `32`.
+
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 131072):
+            The maximum sequence length that this model might ever be used with.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type
+            and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value
+            accordingly.
+            Expected contents:
+                `rope_type` (`str`):
+                    The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
+                    'llama3'], with 'default' being the original RoPE implementation.
+                `factor` (`float`, *optional*):
+                    Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
+                    most scaling types, a `factor` of x will enable the model to handle sequences of length x *
+                    original maximum pre-trained length.
+                `original_max_position_embeddings` (`int`, *optional*):
+                    Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
+                    pretraining.
+                `attention_factor` (`float`, *optional*):
+                    Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
+                    computation. If unspecified, it defaults to value recommended by the implementation, using the
+                    `factor` field to infer the suggested value.
+                `beta_fast` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 32.
+                `beta_slow` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 1.
+                `short_factor` (`list[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to short contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `long_factor` (`list[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to long contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `low_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
+                `high_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        moe_intermediate_size (`int`, *optional*, defaults to 1408):
+            Intermediate size of the routed expert.
+        num_experts_per_tok (`int`, *optional*, defaults to 8):
+            number of experts per token.
+        n_shared_experts (`int`, *optional*, defaults to 1):
+            Number of shared experts.
+        n_routed_experts (`int`, *optional*, defaults to 128):
+            Number of routed experts.
+        routed_scaling_factor (`float`, *optional*, defaults to 1.0):
+            Scaling factor or routed experts.
+        n_group (`int`, *optional*, defaults to 1):
+            Number of groups for routed experts.
+        topk_group (`int`, *optional*, defaults to 1):
+            Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
+        first_k_dense_replace (`int`, *optional*, defaults to 1):
+            Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
+                                                            \--k dense layers--/
+        norm_topk_prob (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the topk probabilities.
+        use_qk_norm (`bool`, *optional*, defaults to `False`):
+            Whether to use query-key normalization in the attention
+    ```python
+    >>> from transformers import Glm4MoeModel, Glm4MoeConfig
+
+    >>> # Initializing a Glm4Moe style configuration
+    >>> configuration = Glm4MoeConfig()
+
+    >>> # Initializing a model from the GLM-4-MOE-100B-A10B style configuration
+    >>> model = Glm4MoeModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "glm4_shared_moe"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    # Default tensor parallel plan for base model `Glm4Moe`
+    base_model_tp_plan = {
+        "layers.*.self_attn.q_proj": "colwise",
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.mlp.experts.*.gate_proj": "colwise",
+        "layers.*.mlp.experts.*.up_proj": "colwise",
+        "layers.*.mlp.experts.*.down_proj": "rowwise",
+        "layers.*.mlp.gate_proj": "colwise",
+        "layers.*.mlp.up_proj": "colwise",
+        "layers.*.mlp.down_proj": "rowwise",
+    }
+    base_model_pp_plan = {
+        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+        "norm": (["hidden_states"], ["hidden_states"]),
+    }
+
+    def __init__(
+        self,
+        vocab_size=151552,
+        hidden_size=4096,
+        intermediate_size=10944,
+        num_hidden_layers=46,
+        num_attention_heads=96,
+        partial_rotary_factor=0.5,
+        num_key_value_heads=8,
+        hidden_act="silu",
+        max_position_embeddings=131072,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        attention_dropout=0.0,
+        moe_intermediate_size=1408,
+        num_experts_per_tok=8,
+        n_shared_experts=1,
+        n_routed_experts=128,
+        routed_scaling_factor=1.0,
+        n_group=1,
+        topk_group=1,
+        first_k_dense_replace=1,
+        norm_topk_prob=True,
+        use_qk_norm=False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.partial_rotary_factor = partial_rotary_factor
+
+        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.rope_scaling = rope_scaling
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        # Validate the correctness of rotary position embeddings parameters
+        # BC: if there is a 'type' field, move it to 'rope_type'.
+        if self.rope_scaling is not None and "type" in self.rope_scaling:
+            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
+        rope_config_validation(self)
+
+        # MoE arguments
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_experts_per_tok = num_experts_per_tok
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.n_shared_experts = n_shared_experts
+        self.n_routed_experts = n_routed_experts
+        self.routed_scaling_factor = routed_scaling_factor
+        self.first_k_dense_replace = first_k_dense_replace
+        self.norm_topk_prob = norm_topk_prob
+        self.use_qk_norm = use_qk_norm
+
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+
+__all__ = ["Glm4SharedMoeConfig"]
diff --git a/convert_hf_to_scm.py b/convert_hf_to_scm.py
new file mode 100644
index 0000000000000000000000000000000000000000..1fd773a856db911eb80eaab8ad5ad4b4886c5233
--- /dev/null
+++ b/convert_hf_to_scm.py
@@ -0,0 +1,126 @@
+import glob
+import re
+import shutil
+import sys
+
+import accelerate
+import torch
+from safetensors import safe_open
+from configuration_glm4_shared_moe import Glm4SharedMoeConfig
+from modeling_glm4_shared_moe import Glm4SharedMoeForCausalLM
+from transformers.models.glm4_moe.configuration_glm4_moe import Glm4MoeConfig
+
+input_model = sys.argv[1]
+output_model_path = sys.argv[2]
+
+auto_map = {
+    "AutoConfig": "configuration_glm4_shared_moe.Glm4SharedMoeConfig",
+    "AutoModel": "modeling_glm4_shared_moe.Glm4SharedMoeModel",
+    "AutoModelForCausalLM": "modeling_glm4_shared_moe.Glm4SharedMoeForCausalLM"
+}
+
+cfg_standard_moe = Glm4MoeConfig.from_pretrained(input_model)
+cfg_shared_moe = Glm4SharedMoeConfig(
+    auto_map=auto_map,
+    vocab_size=cfg_standard_moe.vocab_size,
+    hidden_size=cfg_standard_moe.hidden_size,
+    intermediate_size=cfg_standard_moe.intermediate_size,
+    num_hidden_layers=cfg_standard_moe.num_hidden_layers,
+    num_attention_heads=cfg_standard_moe.num_attention_heads,
+    num_key_value_heads=cfg_standard_moe.num_key_value_heads,
+    hidden_act=cfg_standard_moe.hidden_act,
+    max_position_embeddings=cfg_standard_moe.max_position_embeddings,
+    initializer_range=cfg_standard_moe.initializer_range,
+    rms_norm_eps=cfg_standard_moe.rms_norm_eps,
+    use_cache=cfg_standard_moe.use_cache,
+    tie_word_embeddings=cfg_standard_moe.tie_word_embeddings,
+    rope_theta=cfg_standard_moe.rope_theta,
+    rope_scaling=cfg_standard_moe.rope_scaling,
+    attention_bias=cfg_standard_moe.attention_bias,
+    attention_dropout=cfg_standard_moe.attention_dropout,
+    moe_intermediate_size=cfg_standard_moe.moe_intermediate_size,
+    num_experts_per_tok=cfg_standard_moe.num_experts_per_tok,
+    n_routed_experts=cfg_standard_moe.n_routed_experts,
+    n_shared_experts=cfg_standard_moe.n_shared_experts,
+    norm_topk_prob=cfg_standard_moe.norm_topk_prob,
+    head_dim=cfg_standard_moe.head_dim,
+    pad_token_id=cfg_standard_moe.pad_token_id,
+    eos_token_id=cfg_standard_moe.eos_token_id,
+    routed_scaling_factor=cfg_standard_moe.routed_scaling_factor,
+    first_k_dense_replace=cfg_standard_moe.first_k_dense_replace,
+    num_nextn_predict_layers=cfg_standard_moe.num_nextn_predict_layers,
+)
+
+num_experts = cfg_standard_moe.n_routed_experts
+
+with accelerate.init_empty_weights():
+    model_shared_moe = Glm4SharedMoeForCausalLM(cfg_shared_moe)
+
+model_shared_moe = model_shared_moe.to(torch.bfloat16)
+new_state_dict = {}
+pattern = f"{input_model}/model-*-of-*.safetensors"
+files = sorted(glob.glob(pattern))
+
+if len(files) == 0:
+    raise FileNotFoundError
+tensors = {}
+
+for file_path in files:
+    print(f"processing {file_path}")
+    with safe_open(file_path, framework="pt", device="cpu") as f:
+        for key in f.keys():
+            tensor = f.get_tensor(key)
+            tensors[key] = tensor
+
+for key in tensors:
+    try:
+        layer_num = int(re.search(r"\d+", key).group())
+        if layer_num >= cfg_standard_moe.num_hidden_layers:
+            continue
+    except:
+        pass
+    if "experts" not in key or "shared_experts" in key:
+        new_state_dict[key] = tensors[key]
+    elif "experts.0" in key:
+        layer_num = int(re.search(r"\d+", key).group())
+        new_state_dict[
+            f"model.layers.{layer_num}.mlp.moe_mlp.output_experts.weight"
+        ] = torch.stack(
+            [
+                tensors[f"model.layers.{layer_num}.mlp.experts.{i}.down_proj.weight"]
+                for i in range(num_experts)
+            ]
+        )
+        new_state_dict[f"model.layers.{layer_num}.mlp.moe_mlp.experts.weight"] = (
+            torch.stack(
+                [
+                    torch.cat(
+                        [
+                            tensors[
+                                f"model.layers.{layer_num}.mlp.experts.{i}.up_proj.weight"
+                            ],
+                            tensors[
+                                f"model.layers.{layer_num}.mlp.experts.{i}.gate_proj.weight"
+                            ],
+                        ],
+                        dim=0,
+                    )
+                    for i in range(num_experts)
+                ]
+            )
+        )
+model_shared_moe.load_state_dict(new_state_dict, strict=True, assign=True)
+model_shared_moe.save_pretrained(output_model_path)
+cfg_shared_moe.save_pretrained(output_model_path)
+
+
+shutil.copy(
+    "modeling_glm4_shared_moe.py",
+    output_model_path + "/" + "modeling_glm4_shared_moe.py",
+)
+shutil.copy(
+    "configuration_glm4_shared_moe.py",
+    output_model_path + "/" + "configuration_glm4_shared_moe.py",
+)
+for i in ["chat_template.jinja", "tokenizer_config.json", "tokenizer.json"]:
+    shutil.copy(input_model + "/" + i, output_model_path + "/" + i)
diff --git a/generation_config.json b/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4fdc8b42f40332520174dfd60584ddfe7d68f6a2
--- /dev/null
+++ b/generation_config.json
@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "eos_token_id": [
+    151329,
+    151336,
+    151338
+  ],
+  "pad_token_id": 151329,
+  "transformers_version": "4.54.1"
+}
diff --git a/model-00001-of-00046.safetensors b/model-00001-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..437ecfbaac0d65b8f057de19ed6e773b969db556
--- /dev/null
+++ b/model-00001-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d45776392254b275a24e244328e0f6cbce58a31dae011c96560f7c656fcd1c2e
+size 4899547528
diff --git a/model-00002-of-00046.safetensors b/model-00002-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b1c08364f91fb996026a10606c364e376eb0a591
--- /dev/null
+++ b/model-00002-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f156975edce2bebc2424265823b7206ef4d01942a5b287060b21072fff5f630
+size 4682987888
diff --git a/model-00003-of-00046.safetensors b/model-00003-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0481ed1cfafd7cee9c43c089bff4c05491a7fe67
--- /dev/null
+++ b/model-00003-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:216f2d894bcf6c49d8fec4d23ce347c5737828a7f56c9fcce8fdd5b8f09a07e4
+size 4682987888
diff --git a/model-00004-of-00046.safetensors b/model-00004-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..116efbe446d50a571e21c292e620515518bcb5ab
--- /dev/null
+++ b/model-00004-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4ebd54e8e5b07c3a92c428d86167908e43df3238d0787d0605bbc647cfa3807d
+size 4682987888
diff --git a/model-00005-of-00046.safetensors b/model-00005-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7f2cd384efba37ed99a069198b6e6bd2e1b1dbea
--- /dev/null
+++ b/model-00005-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:800c75f575211194ceb2459d75711b2e52a5ce9c368e39e2fc6842552d5bbeec
+size 4682987888
diff --git a/model-00006-of-00046.safetensors b/model-00006-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5b33cbae923e9cd7defdd2f8b71220a0bf78e1a0
--- /dev/null
+++ b/model-00006-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0555893367e37ff2fff426272903404743574de456c20d72a6eb01bb76f9638
+size 4682987888
diff --git a/model-00007-of-00046.safetensors b/model-00007-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4d0d03498ef5cdad0b3ead178fc8409d060ba0d5
--- /dev/null
+++ b/model-00007-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0d3b02e2d94d6f706ecb171b15c7d31f0a8ca8127ebf65ce1475d1a9adae248
+size 4682987888
diff --git a/model-00008-of-00046.safetensors b/model-00008-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2ee187507171b99f7ad0a342a4c4e18b8217573d
--- /dev/null
+++ b/model-00008-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:263f71ee54b5c440b7277922fb1d2366fb5c573bae182c4ab7fb0b6ac0369162
+size 4682987888
diff --git a/model-00009-of-00046.safetensors b/model-00009-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5e0369f9142f807729f419e6074fbe66d56464ef
--- /dev/null
+++ b/model-00009-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b825805e3185f279ee194ab2c8261abf547ed01d40971dc1758578da0e43743b
+size 4682987888
diff --git a/model-00010-of-00046.safetensors b/model-00010-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fe644851bb86e9bc699142c4cd4d04221cad9bb3
--- /dev/null
+++ b/model-00010-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae61902a0b3805a5b4fbe99cc28280bdbc897d13b10549c784c8d7f5bdb9bdc3
+size 4682987912
diff --git a/model-00011-of-00046.safetensors b/model-00011-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1d66c858c31c157cfacc90d87567fb1b9559c4ad
--- /dev/null
+++ b/model-00011-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95ee704695042ffb44b599b327f51f3a630cbec23831ffa25003386b32990926
+size 4682987904
diff --git a/model-00012-of-00046.safetensors b/model-00012-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..feec884c3d7f8e54decea9ea65647cc1bd3b33fc
--- /dev/null
+++ b/model-00012-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:445e13c51c454d556b4390de5431f49afd128d1eed971b4ac29676bb67e15295
+size 4682987904
diff --git a/model-00013-of-00046.safetensors b/model-00013-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..59ee3687185dbc34a99eebda1c731607c77301d5
--- /dev/null
+++ b/model-00013-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7326b4961c34c17f812e5e19ee55ab1fd694fc68f8e98819e96e862999b5d442
+size 4682987904
diff --git a/model-00014-of-00046.safetensors b/model-00014-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f30cee8e8ab0d4f6a2dd8f763f200b33cec97d67
--- /dev/null
+++ b/model-00014-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfc8befb7da9721cb8528fad81e012ab89b2828fede9a283a7ed20a4332e9746
+size 4682987904
diff --git a/model-00015-of-00046.safetensors b/model-00015-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c144cb3be742132b7f4dcf1c037701d9dd0565a1
--- /dev/null
+++ b/model-00015-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22015ca4df121eb8cfc5acd01354fc369fcf947f51b27652b6d04bba9cb9696b
+size 4682987904
diff --git a/model-00016-of-00046.safetensors b/model-00016-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2373ea8f82febab98705da463858e8935e1919e7
--- /dev/null
+++ b/model-00016-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:780a49d1d3e54e7b37b45b9efc0d02eec3140cefe4845bbd077d2c4a71039c59
+size 4682987904
diff --git a/model-00017-of-00046.safetensors b/model-00017-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..de97ede637e3a070a889992ff111bf5ec622b73c
--- /dev/null
+++ b/model-00017-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0be7537161cee75fbff97df03d9e585104f976b31fb73a098f8975b33b3111c
+size 4682987904
diff --git a/model-00018-of-00046.safetensors b/model-00018-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9f881688f0fc57a63961e6eaf70acefa064b43c8
--- /dev/null
+++ b/model-00018-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5528044fab3e165d495b05dfe2076fbaf06457083eecf508b28cee9b3b4ed5f1
+size 4682987904
diff --git a/model-00019-of-00046.safetensors b/model-00019-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9f511c6d385e46fb0c4b8ebd4a1cc25099b9fd02
--- /dev/null
+++ b/model-00019-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f750c39a564577349c7705ec95cd4aeb73624d100df86ad95f7b8b350c2e660
+size 4682987904
diff --git a/model-00020-of-00046.safetensors b/model-00020-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1b2a3f23c4d3b29ff5b782ee85f4298d2fa824c7
--- /dev/null
+++ b/model-00020-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ce649f5393c57a81707a26439f82d89db5df2bb0e4ffc7b2874192e7a0735c1
+size 4682987904
diff --git a/model-00021-of-00046.safetensors b/model-00021-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4eb2e6236449b9503f00468c7dd54368b67836fb
--- /dev/null
+++ b/model-00021-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c67d951bdc8567d6e03d2ca9cd69779a96afeb090edb8e257d78fcd0ffcb33e5
+size 4682987904
diff --git a/model-00022-of-00046.safetensors b/model-00022-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bd734e30689e7af1dc4cde2b613193a526d39fd9
--- /dev/null
+++ b/model-00022-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1ba8741ebe985e32c17ef26e06526223d306d73ac15f7a6600c24a22e9111ac
+size 4682987904
diff --git a/model-00023-of-00046.safetensors b/model-00023-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b910a11c3f71194a6d9ef2e744f1ee2c35fa037a
--- /dev/null
+++ b/model-00023-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7de865761fae8c7e3c92fc7e7af3e7ba14a8846c4a97944e410181fc451d2d7
+size 4682987904
diff --git a/model-00024-of-00046.safetensors b/model-00024-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2f65920443109e2e32358b267d73f2e5c681f30b
--- /dev/null
+++ b/model-00024-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d1f833392d772aab9f079972ace909b94e28c5c895668f153e35bb9eb100fe8
+size 4682987904
diff --git a/model-00025-of-00046.safetensors b/model-00025-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e40bd51755d22ececd712dd86985d8a5a23d9258
--- /dev/null
+++ b/model-00025-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b734d236f3f095c50c1d1ac72dc8d02ab6a6fd1e90220663a23ea39c1ae591f
+size 4682987904
diff --git a/model-00026-of-00046.safetensors b/model-00026-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b6f1671bf19930ffa5a5ba36ac30f3f944e1d36c
--- /dev/null
+++ b/model-00026-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:282e19852b5b56b728be3aee8b8f41d516347c4f55722eefa59ac7dc270adfe2
+size 4682987904
diff --git a/model-00027-of-00046.safetensors b/model-00027-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..2e6a0fedd725ff7dd7baac2bed4fbea51b175993
--- /dev/null
+++ b/model-00027-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ca59588d218b55bb4c7cb765788c668a05f1a7acaa8eb50b1864c915b004f96
+size 4682987904
diff --git a/model-00028-of-00046.safetensors b/model-00028-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e9588c4d7e40facbeeafa3ef05bfea1242707acb
--- /dev/null
+++ b/model-00028-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28c33a4dce7f4bab03c1816cb0548e3dd18dafd71a990837b5e943649da04b91
+size 4682987904
diff --git a/model-00029-of-00046.safetensors b/model-00029-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..762bd997b6dc68f947d6503a28daf921bb9a2728
--- /dev/null
+++ b/model-00029-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f10451b115af4d6c831155d8abce79905c06f5bd42db9551224d2260c2573c37
+size 4682987904
diff --git a/model-00030-of-00046.safetensors b/model-00030-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..754db4b57dd4f9c19e6262a83b4cb178bb9f435b
--- /dev/null
+++ b/model-00030-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07c29107055802d27415e2220abf4ac727202bc3abdf8e607c75b4603e8185c2
+size 4682987904
diff --git a/model-00031-of-00046.safetensors b/model-00031-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fe2c8e6d43fd666396ed22cc8b03256bb117362e
--- /dev/null
+++ b/model-00031-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3b8c0fabd37def5075c9cc4510a78f181ae12a4b15c0ea0d9b0907ca8512fce
+size 4682987904
diff --git a/model-00032-of-00046.safetensors b/model-00032-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bfa2570700764c5b5e13d1bfea3ec94c2b15513d
--- /dev/null
+++ b/model-00032-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ec8480d163b94b1f321309a505ac1d3ea48b3f1fa86ba9ad7df0b3c9112ef36
+size 4682987904
diff --git a/model-00033-of-00046.safetensors b/model-00033-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3d3687a8dfd88cef03460f680eb0e81d1548ef06
--- /dev/null
+++ b/model-00033-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8519f57ab1fbd15324f8939503c61328ae8a60940e892192ca19a927f85bb7ea
+size 4682987904
diff --git a/model-00034-of-00046.safetensors b/model-00034-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dc2d0b6cfd9fd399ce79a9da25ceb80088b90c4d
--- /dev/null
+++ b/model-00034-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c8076078d3ccdbdec4d375ad01e65b8c741b35e83bf772f52667b5facf492d3
+size 4682987904
diff --git a/model-00035-of-00046.safetensors b/model-00035-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b436990b9b0377f7c99208a4289211386c2e259f
--- /dev/null
+++ b/model-00035-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d64eca561f4471bdb4438150f7a67b147d6acdc1b028cc0b8770ea979c390136
+size 4682987904
diff --git a/model-00036-of-00046.safetensors b/model-00036-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9ccac20ae344cea383169e0e4f711e4abc2eca2e
--- /dev/null
+++ b/model-00036-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79bbc64f7704ff01f30bb3b9038c5a23bb9d068a44577616db7e8854faee8dd3
+size 4682987904
diff --git a/model-00037-of-00046.safetensors b/model-00037-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..dfac5bed87a8ea4f6f7db54b53e81e4e8cb8a1fe
--- /dev/null
+++ b/model-00037-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af65b309aa1d7e7e9bec62a56dd40544b19e595eae1071510a66817ff5c2bbef
+size 4682987904
diff --git a/model-00038-of-00046.safetensors b/model-00038-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3324566793412a309258294194235296c8ab16da
--- /dev/null
+++ b/model-00038-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1607ec1d64f4b46a26770af361e5be819f919d2aab0b0f4f82fac3753f3ff0a
+size 4682987904
diff --git a/model-00039-of-00046.safetensors b/model-00039-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0ac6bd98f70eaf3940d289b15bb60ac362fc6570
--- /dev/null
+++ b/model-00039-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca51aec4dc1b1b456ee324edfa7f4b51c0f74fa80c6458111f6523408adb5dcc
+size 4682987904
diff --git a/model-00040-of-00046.safetensors b/model-00040-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..da5e8fbc0b0ff9faa7c96f204334544a4b349117
--- /dev/null
+++ b/model-00040-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcea05440633d1a87c820d1f31c95bd80b372d6937560422153a224dcc966738
+size 4682987904
diff --git a/model-00041-of-00046.safetensors b/model-00041-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..13222fac58ee5e19b92bb197b2740a1c9ce6899c
--- /dev/null
+++ b/model-00041-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16d275875245d7c920608dba86e718c3d351a92e4a02d48973647748087e7f75
+size 4682987904
diff --git a/model-00042-of-00046.safetensors b/model-00042-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e5be5817cf2908acf2214c37a615ec8e82f2f021
--- /dev/null
+++ b/model-00042-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37547c91b22403a1044f3cfc7dc227aae89d483b8103292da9038ace6d50c539
+size 4682987904
diff --git a/model-00043-of-00046.safetensors b/model-00043-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..36e7e0ec3c6e95a6a69503ce8b5d042b76216712
--- /dev/null
+++ b/model-00043-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a326957d379158303e37b2ea9f51db985fd0f56d3b5d5343736ad20d943bc84d
+size 4682987904
diff --git a/model-00044-of-00046.safetensors b/model-00044-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..28a48a9ba1914e5c54c8513ca7072367c72e9fbf
--- /dev/null
+++ b/model-00044-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46ea1d2d23665d8be7493be796b10b3fd2b9ae6193c935f8fd4f6f6489733c4e
+size 4682987904
diff --git a/model-00045-of-00046.safetensors b/model-00045-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9b56dc9f35936e4f385991b6e9efa8e1c81fd934
--- /dev/null
+++ b/model-00045-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a3661c735da2ec4079933445b8a001444c760092d3e31d387d4ec7e2c0e9512
+size 4682987904
diff --git a/model-00046-of-00046.safetensors b/model-00046-of-00046.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..29f5b4001500120e3c4ec7c6acadeefb2ec33f77
--- /dev/null
+++ b/model-00046-of-00046.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fae46a9168075e1566974bfa1d7dd2fab9edb477edc6b4965564475ca192c7a
+size 2753586856
diff --git a/model.safetensors.index.json b/model.safetensors.index.json
new file mode 100644
index 0000000000000000000000000000000000000000..543b9fbe6df5a7e8c9e1ae59dcbee90a2c7f8b46
--- /dev/null
+++ b/model.safetensors.index.json
@@ -0,0 +1,743 @@
+{
+  "metadata": {
+    "total_parameters": 106852245504,
+    "total_size": 213704514048
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00046-of-00046.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.k_proj.bias": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.q_proj.bias": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.v_proj.bias": "model-00001-of-00046.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.mlp.gate.e_score_correction_bias": "model-00002-of-00046.safetensors",
+    "model.layers.1.mlp.gate.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.mlp.moe_mlp.experts.weight": "model-00001-of-00046.safetensors",
+    "model.layers.1.mlp.moe_mlp.output_experts.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.mlp.shared_experts.down_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.mlp.shared_experts.gate_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.mlp.shared_experts.up_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00002-of-00046.safetensors",
+    "model.layers.1.self_attn.k_proj.bias": "model-00001-of-00046.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.1.self_attn.q_proj.bias": "model-00001-of-00046.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.1.self_attn.v_proj.bias": "model-00001-of-00046.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00046.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.mlp.gate.e_score_correction_bias": "model-00011-of-00046.safetensors",
+    "model.layers.10.mlp.gate.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.mlp.moe_mlp.experts.weight": "model-00010-of-00046.safetensors",
+    "model.layers.10.mlp.moe_mlp.output_experts.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.mlp.shared_experts.down_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.mlp.shared_experts.gate_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.mlp.shared_experts.up_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00011-of-00046.safetensors",
+    "model.layers.10.self_attn.k_proj.bias": "model-00010-of-00046.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.10.self_attn.q_proj.bias": "model-00010-of-00046.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.10.self_attn.v_proj.bias": "model-00010-of-00046.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.mlp.gate.e_score_correction_bias": "model-00012-of-00046.safetensors",
+    "model.layers.11.mlp.gate.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.mlp.moe_mlp.experts.weight": "model-00011-of-00046.safetensors",
+    "model.layers.11.mlp.moe_mlp.output_experts.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.mlp.shared_experts.down_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.mlp.shared_experts.gate_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.mlp.shared_experts.up_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00012-of-00046.safetensors",
+    "model.layers.11.self_attn.k_proj.bias": "model-00011-of-00046.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.11.self_attn.q_proj.bias": "model-00011-of-00046.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.11.self_attn.v_proj.bias": "model-00011-of-00046.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00011-of-00046.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.mlp.gate.e_score_correction_bias": "model-00013-of-00046.safetensors",
+    "model.layers.12.mlp.gate.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.mlp.moe_mlp.experts.weight": "model-00012-of-00046.safetensors",
+    "model.layers.12.mlp.moe_mlp.output_experts.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.mlp.shared_experts.down_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.mlp.shared_experts.gate_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.mlp.shared_experts.up_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00013-of-00046.safetensors",
+    "model.layers.12.self_attn.k_proj.bias": "model-00012-of-00046.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.12.self_attn.q_proj.bias": "model-00012-of-00046.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.12.self_attn.v_proj.bias": "model-00012-of-00046.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00012-of-00046.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.mlp.gate.e_score_correction_bias": "model-00014-of-00046.safetensors",
+    "model.layers.13.mlp.gate.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.mlp.moe_mlp.experts.weight": "model-00013-of-00046.safetensors",
+    "model.layers.13.mlp.moe_mlp.output_experts.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.mlp.shared_experts.down_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.mlp.shared_experts.gate_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.mlp.shared_experts.up_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00014-of-00046.safetensors",
+    "model.layers.13.self_attn.k_proj.bias": "model-00013-of-00046.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.13.self_attn.q_proj.bias": "model-00013-of-00046.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.13.self_attn.v_proj.bias": "model-00013-of-00046.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00013-of-00046.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.mlp.gate.e_score_correction_bias": "model-00015-of-00046.safetensors",
+    "model.layers.14.mlp.gate.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.mlp.moe_mlp.experts.weight": "model-00014-of-00046.safetensors",
+    "model.layers.14.mlp.moe_mlp.output_experts.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.mlp.shared_experts.down_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.mlp.shared_experts.gate_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.mlp.shared_experts.up_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00015-of-00046.safetensors",
+    "model.layers.14.self_attn.k_proj.bias": "model-00014-of-00046.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.14.self_attn.q_proj.bias": "model-00014-of-00046.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.14.self_attn.v_proj.bias": "model-00014-of-00046.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00014-of-00046.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.mlp.gate.e_score_correction_bias": "model-00016-of-00046.safetensors",
+    "model.layers.15.mlp.gate.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.mlp.moe_mlp.experts.weight": "model-00015-of-00046.safetensors",
+    "model.layers.15.mlp.moe_mlp.output_experts.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.mlp.shared_experts.down_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.mlp.shared_experts.gate_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.mlp.shared_experts.up_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00016-of-00046.safetensors",
+    "model.layers.15.self_attn.k_proj.bias": "model-00015-of-00046.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.15.self_attn.q_proj.bias": "model-00015-of-00046.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.15.self_attn.v_proj.bias": "model-00015-of-00046.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00015-of-00046.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.mlp.gate.e_score_correction_bias": "model-00017-of-00046.safetensors",
+    "model.layers.16.mlp.gate.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.mlp.moe_mlp.experts.weight": "model-00016-of-00046.safetensors",
+    "model.layers.16.mlp.moe_mlp.output_experts.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.mlp.shared_experts.down_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.mlp.shared_experts.gate_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.mlp.shared_experts.up_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00017-of-00046.safetensors",
+    "model.layers.16.self_attn.k_proj.bias": "model-00016-of-00046.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.16.self_attn.q_proj.bias": "model-00016-of-00046.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.16.self_attn.v_proj.bias": "model-00016-of-00046.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00016-of-00046.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.mlp.gate.e_score_correction_bias": "model-00018-of-00046.safetensors",
+    "model.layers.17.mlp.gate.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.mlp.moe_mlp.experts.weight": "model-00017-of-00046.safetensors",
+    "model.layers.17.mlp.moe_mlp.output_experts.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.mlp.shared_experts.down_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.mlp.shared_experts.gate_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.mlp.shared_experts.up_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00018-of-00046.safetensors",
+    "model.layers.17.self_attn.k_proj.bias": "model-00017-of-00046.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.17.self_attn.q_proj.bias": "model-00017-of-00046.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.17.self_attn.v_proj.bias": "model-00017-of-00046.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00017-of-00046.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.mlp.gate.e_score_correction_bias": "model-00019-of-00046.safetensors",
+    "model.layers.18.mlp.gate.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.mlp.moe_mlp.experts.weight": "model-00018-of-00046.safetensors",
+    "model.layers.18.mlp.moe_mlp.output_experts.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.mlp.shared_experts.down_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.mlp.shared_experts.gate_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.mlp.shared_experts.up_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00019-of-00046.safetensors",
+    "model.layers.18.self_attn.k_proj.bias": "model-00018-of-00046.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.18.self_attn.q_proj.bias": "model-00018-of-00046.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.18.self_attn.v_proj.bias": "model-00018-of-00046.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00018-of-00046.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.mlp.gate.e_score_correction_bias": "model-00020-of-00046.safetensors",
+    "model.layers.19.mlp.gate.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.mlp.moe_mlp.experts.weight": "model-00019-of-00046.safetensors",
+    "model.layers.19.mlp.moe_mlp.output_experts.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.mlp.shared_experts.down_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.mlp.shared_experts.gate_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.mlp.shared_experts.up_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00020-of-00046.safetensors",
+    "model.layers.19.self_attn.k_proj.bias": "model-00019-of-00046.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.19.self_attn.q_proj.bias": "model-00019-of-00046.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.19.self_attn.v_proj.bias": "model-00019-of-00046.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00019-of-00046.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.mlp.gate.e_score_correction_bias": "model-00003-of-00046.safetensors",
+    "model.layers.2.mlp.gate.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.mlp.moe_mlp.experts.weight": "model-00002-of-00046.safetensors",
+    "model.layers.2.mlp.moe_mlp.output_experts.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.mlp.shared_experts.down_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.mlp.shared_experts.gate_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.mlp.shared_experts.up_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00003-of-00046.safetensors",
+    "model.layers.2.self_attn.k_proj.bias": "model-00002-of-00046.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.2.self_attn.q_proj.bias": "model-00002-of-00046.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.2.self_attn.v_proj.bias": "model-00002-of-00046.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00002-of-00046.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.mlp.gate.e_score_correction_bias": "model-00021-of-00046.safetensors",
+    "model.layers.20.mlp.gate.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.mlp.moe_mlp.experts.weight": "model-00020-of-00046.safetensors",
+    "model.layers.20.mlp.moe_mlp.output_experts.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.mlp.shared_experts.down_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.mlp.shared_experts.gate_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.mlp.shared_experts.up_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00021-of-00046.safetensors",
+    "model.layers.20.self_attn.k_proj.bias": "model-00020-of-00046.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.20.self_attn.q_proj.bias": "model-00020-of-00046.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.20.self_attn.v_proj.bias": "model-00020-of-00046.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00020-of-00046.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.mlp.gate.e_score_correction_bias": "model-00022-of-00046.safetensors",
+    "model.layers.21.mlp.gate.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.mlp.moe_mlp.experts.weight": "model-00021-of-00046.safetensors",
+    "model.layers.21.mlp.moe_mlp.output_experts.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.mlp.shared_experts.down_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.mlp.shared_experts.gate_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.mlp.shared_experts.up_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00022-of-00046.safetensors",
+    "model.layers.21.self_attn.k_proj.bias": "model-00021-of-00046.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.21.self_attn.q_proj.bias": "model-00021-of-00046.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.21.self_attn.v_proj.bias": "model-00021-of-00046.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00021-of-00046.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.mlp.gate.e_score_correction_bias": "model-00023-of-00046.safetensors",
+    "model.layers.22.mlp.gate.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.mlp.moe_mlp.experts.weight": "model-00022-of-00046.safetensors",
+    "model.layers.22.mlp.moe_mlp.output_experts.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.mlp.shared_experts.down_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.mlp.shared_experts.gate_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.mlp.shared_experts.up_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00023-of-00046.safetensors",
+    "model.layers.22.self_attn.k_proj.bias": "model-00022-of-00046.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.22.self_attn.q_proj.bias": "model-00022-of-00046.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.22.self_attn.v_proj.bias": "model-00022-of-00046.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00022-of-00046.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.mlp.gate.e_score_correction_bias": "model-00024-of-00046.safetensors",
+    "model.layers.23.mlp.gate.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.mlp.moe_mlp.experts.weight": "model-00023-of-00046.safetensors",
+    "model.layers.23.mlp.moe_mlp.output_experts.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.mlp.shared_experts.down_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.mlp.shared_experts.gate_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.mlp.shared_experts.up_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00024-of-00046.safetensors",
+    "model.layers.23.self_attn.k_proj.bias": "model-00023-of-00046.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.23.self_attn.q_proj.bias": "model-00023-of-00046.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.23.self_attn.v_proj.bias": "model-00023-of-00046.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00023-of-00046.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.mlp.gate.e_score_correction_bias": "model-00025-of-00046.safetensors",
+    "model.layers.24.mlp.gate.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.mlp.moe_mlp.experts.weight": "model-00024-of-00046.safetensors",
+    "model.layers.24.mlp.moe_mlp.output_experts.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.mlp.shared_experts.down_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.mlp.shared_experts.gate_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.mlp.shared_experts.up_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00025-of-00046.safetensors",
+    "model.layers.24.self_attn.k_proj.bias": "model-00024-of-00046.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.24.self_attn.q_proj.bias": "model-00024-of-00046.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.24.self_attn.v_proj.bias": "model-00024-of-00046.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00024-of-00046.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.mlp.gate.e_score_correction_bias": "model-00026-of-00046.safetensors",
+    "model.layers.25.mlp.gate.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.mlp.moe_mlp.experts.weight": "model-00025-of-00046.safetensors",
+    "model.layers.25.mlp.moe_mlp.output_experts.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.mlp.shared_experts.down_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.mlp.shared_experts.gate_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.mlp.shared_experts.up_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00026-of-00046.safetensors",
+    "model.layers.25.self_attn.k_proj.bias": "model-00025-of-00046.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.25.self_attn.q_proj.bias": "model-00025-of-00046.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.25.self_attn.v_proj.bias": "model-00025-of-00046.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00025-of-00046.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.mlp.gate.e_score_correction_bias": "model-00027-of-00046.safetensors",
+    "model.layers.26.mlp.gate.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.mlp.moe_mlp.experts.weight": "model-00026-of-00046.safetensors",
+    "model.layers.26.mlp.moe_mlp.output_experts.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.mlp.shared_experts.down_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.mlp.shared_experts.gate_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.mlp.shared_experts.up_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00027-of-00046.safetensors",
+    "model.layers.26.self_attn.k_proj.bias": "model-00026-of-00046.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.26.self_attn.q_proj.bias": "model-00026-of-00046.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.26.self_attn.v_proj.bias": "model-00026-of-00046.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00026-of-00046.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.mlp.gate.e_score_correction_bias": "model-00028-of-00046.safetensors",
+    "model.layers.27.mlp.gate.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.mlp.moe_mlp.experts.weight": "model-00027-of-00046.safetensors",
+    "model.layers.27.mlp.moe_mlp.output_experts.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.mlp.shared_experts.down_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.mlp.shared_experts.gate_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.mlp.shared_experts.up_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00028-of-00046.safetensors",
+    "model.layers.27.self_attn.k_proj.bias": "model-00027-of-00046.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.27.self_attn.q_proj.bias": "model-00027-of-00046.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.27.self_attn.v_proj.bias": "model-00027-of-00046.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00027-of-00046.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.mlp.gate.e_score_correction_bias": "model-00029-of-00046.safetensors",
+    "model.layers.28.mlp.gate.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.mlp.moe_mlp.experts.weight": "model-00028-of-00046.safetensors",
+    "model.layers.28.mlp.moe_mlp.output_experts.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.mlp.shared_experts.down_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.mlp.shared_experts.gate_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.mlp.shared_experts.up_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00029-of-00046.safetensors",
+    "model.layers.28.self_attn.k_proj.bias": "model-00028-of-00046.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.28.self_attn.q_proj.bias": "model-00028-of-00046.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.28.self_attn.v_proj.bias": "model-00028-of-00046.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00028-of-00046.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.mlp.gate.e_score_correction_bias": "model-00030-of-00046.safetensors",
+    "model.layers.29.mlp.gate.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.mlp.moe_mlp.experts.weight": "model-00029-of-00046.safetensors",
+    "model.layers.29.mlp.moe_mlp.output_experts.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.mlp.shared_experts.down_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.mlp.shared_experts.gate_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.mlp.shared_experts.up_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00030-of-00046.safetensors",
+    "model.layers.29.self_attn.k_proj.bias": "model-00029-of-00046.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.29.self_attn.q_proj.bias": "model-00029-of-00046.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.29.self_attn.v_proj.bias": "model-00029-of-00046.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00029-of-00046.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.mlp.gate.e_score_correction_bias": "model-00004-of-00046.safetensors",
+    "model.layers.3.mlp.gate.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.mlp.moe_mlp.experts.weight": "model-00003-of-00046.safetensors",
+    "model.layers.3.mlp.moe_mlp.output_experts.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.mlp.shared_experts.down_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.mlp.shared_experts.gate_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.mlp.shared_experts.up_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00004-of-00046.safetensors",
+    "model.layers.3.self_attn.k_proj.bias": "model-00003-of-00046.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.3.self_attn.q_proj.bias": "model-00003-of-00046.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.3.self_attn.v_proj.bias": "model-00003-of-00046.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00003-of-00046.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.mlp.gate.e_score_correction_bias": "model-00031-of-00046.safetensors",
+    "model.layers.30.mlp.gate.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.mlp.moe_mlp.experts.weight": "model-00030-of-00046.safetensors",
+    "model.layers.30.mlp.moe_mlp.output_experts.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.mlp.shared_experts.down_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.mlp.shared_experts.gate_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.mlp.shared_experts.up_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00031-of-00046.safetensors",
+    "model.layers.30.self_attn.k_proj.bias": "model-00030-of-00046.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.30.self_attn.q_proj.bias": "model-00030-of-00046.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.30.self_attn.v_proj.bias": "model-00030-of-00046.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00030-of-00046.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.mlp.gate.e_score_correction_bias": "model-00032-of-00046.safetensors",
+    "model.layers.31.mlp.gate.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.mlp.moe_mlp.experts.weight": "model-00031-of-00046.safetensors",
+    "model.layers.31.mlp.moe_mlp.output_experts.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.mlp.shared_experts.down_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.mlp.shared_experts.gate_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.mlp.shared_experts.up_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00032-of-00046.safetensors",
+    "model.layers.31.self_attn.k_proj.bias": "model-00031-of-00046.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.31.self_attn.q_proj.bias": "model-00031-of-00046.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.31.self_attn.v_proj.bias": "model-00031-of-00046.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00031-of-00046.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.mlp.gate.e_score_correction_bias": "model-00033-of-00046.safetensors",
+    "model.layers.32.mlp.gate.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.mlp.moe_mlp.experts.weight": "model-00032-of-00046.safetensors",
+    "model.layers.32.mlp.moe_mlp.output_experts.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.mlp.shared_experts.down_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.mlp.shared_experts.gate_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.mlp.shared_experts.up_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00033-of-00046.safetensors",
+    "model.layers.32.self_attn.k_proj.bias": "model-00032-of-00046.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.32.self_attn.q_proj.bias": "model-00032-of-00046.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.32.self_attn.v_proj.bias": "model-00032-of-00046.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00032-of-00046.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.mlp.gate.e_score_correction_bias": "model-00034-of-00046.safetensors",
+    "model.layers.33.mlp.gate.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.mlp.moe_mlp.experts.weight": "model-00033-of-00046.safetensors",
+    "model.layers.33.mlp.moe_mlp.output_experts.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.mlp.shared_experts.down_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.mlp.shared_experts.gate_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.mlp.shared_experts.up_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00034-of-00046.safetensors",
+    "model.layers.33.self_attn.k_proj.bias": "model-00033-of-00046.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.33.self_attn.q_proj.bias": "model-00033-of-00046.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.33.self_attn.v_proj.bias": "model-00033-of-00046.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00033-of-00046.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.mlp.gate.e_score_correction_bias": "model-00035-of-00046.safetensors",
+    "model.layers.34.mlp.gate.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.mlp.moe_mlp.experts.weight": "model-00034-of-00046.safetensors",
+    "model.layers.34.mlp.moe_mlp.output_experts.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.mlp.shared_experts.down_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.mlp.shared_experts.gate_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.mlp.shared_experts.up_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00035-of-00046.safetensors",
+    "model.layers.34.self_attn.k_proj.bias": "model-00034-of-00046.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.34.self_attn.q_proj.bias": "model-00034-of-00046.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.34.self_attn.v_proj.bias": "model-00034-of-00046.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00034-of-00046.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.mlp.gate.e_score_correction_bias": "model-00036-of-00046.safetensors",
+    "model.layers.35.mlp.gate.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.mlp.moe_mlp.experts.weight": "model-00035-of-00046.safetensors",
+    "model.layers.35.mlp.moe_mlp.output_experts.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.mlp.shared_experts.down_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.mlp.shared_experts.gate_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.mlp.shared_experts.up_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00036-of-00046.safetensors",
+    "model.layers.35.self_attn.k_proj.bias": "model-00035-of-00046.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.35.self_attn.q_proj.bias": "model-00035-of-00046.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.35.self_attn.v_proj.bias": "model-00035-of-00046.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00035-of-00046.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.mlp.gate.e_score_correction_bias": "model-00037-of-00046.safetensors",
+    "model.layers.36.mlp.gate.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.mlp.moe_mlp.experts.weight": "model-00036-of-00046.safetensors",
+    "model.layers.36.mlp.moe_mlp.output_experts.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.mlp.shared_experts.down_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.mlp.shared_experts.gate_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.mlp.shared_experts.up_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00037-of-00046.safetensors",
+    "model.layers.36.self_attn.k_proj.bias": "model-00036-of-00046.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.36.self_attn.q_proj.bias": "model-00036-of-00046.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.36.self_attn.v_proj.bias": "model-00036-of-00046.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00036-of-00046.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.mlp.gate.e_score_correction_bias": "model-00038-of-00046.safetensors",
+    "model.layers.37.mlp.gate.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.mlp.moe_mlp.experts.weight": "model-00037-of-00046.safetensors",
+    "model.layers.37.mlp.moe_mlp.output_experts.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.mlp.shared_experts.down_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.mlp.shared_experts.gate_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.mlp.shared_experts.up_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00038-of-00046.safetensors",
+    "model.layers.37.self_attn.k_proj.bias": "model-00037-of-00046.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.37.self_attn.q_proj.bias": "model-00037-of-00046.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.37.self_attn.v_proj.bias": "model-00037-of-00046.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00037-of-00046.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.mlp.gate.e_score_correction_bias": "model-00039-of-00046.safetensors",
+    "model.layers.38.mlp.gate.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.mlp.moe_mlp.experts.weight": "model-00038-of-00046.safetensors",
+    "model.layers.38.mlp.moe_mlp.output_experts.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.mlp.shared_experts.down_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.mlp.shared_experts.gate_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.mlp.shared_experts.up_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00039-of-00046.safetensors",
+    "model.layers.38.self_attn.k_proj.bias": "model-00038-of-00046.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.38.self_attn.q_proj.bias": "model-00038-of-00046.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.38.self_attn.v_proj.bias": "model-00038-of-00046.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00038-of-00046.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.mlp.gate.e_score_correction_bias": "model-00040-of-00046.safetensors",
+    "model.layers.39.mlp.gate.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.mlp.moe_mlp.experts.weight": "model-00039-of-00046.safetensors",
+    "model.layers.39.mlp.moe_mlp.output_experts.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.mlp.shared_experts.down_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.mlp.shared_experts.gate_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.mlp.shared_experts.up_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00040-of-00046.safetensors",
+    "model.layers.39.self_attn.k_proj.bias": "model-00039-of-00046.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.39.self_attn.q_proj.bias": "model-00039-of-00046.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.39.self_attn.v_proj.bias": "model-00039-of-00046.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00039-of-00046.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.mlp.gate.e_score_correction_bias": "model-00005-of-00046.safetensors",
+    "model.layers.4.mlp.gate.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.mlp.moe_mlp.experts.weight": "model-00004-of-00046.safetensors",
+    "model.layers.4.mlp.moe_mlp.output_experts.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.mlp.shared_experts.down_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.mlp.shared_experts.gate_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.mlp.shared_experts.up_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00005-of-00046.safetensors",
+    "model.layers.4.self_attn.k_proj.bias": "model-00004-of-00046.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.4.self_attn.q_proj.bias": "model-00004-of-00046.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.4.self_attn.v_proj.bias": "model-00004-of-00046.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00004-of-00046.safetensors",
+    "model.layers.40.input_layernorm.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.mlp.gate.e_score_correction_bias": "model-00041-of-00046.safetensors",
+    "model.layers.40.mlp.gate.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.mlp.moe_mlp.experts.weight": "model-00040-of-00046.safetensors",
+    "model.layers.40.mlp.moe_mlp.output_experts.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.mlp.shared_experts.down_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.mlp.shared_experts.gate_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.mlp.shared_experts.up_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.post_attention_layernorm.weight": "model-00041-of-00046.safetensors",
+    "model.layers.40.self_attn.k_proj.bias": "model-00040-of-00046.safetensors",
+    "model.layers.40.self_attn.k_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.40.self_attn.o_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.40.self_attn.q_proj.bias": "model-00040-of-00046.safetensors",
+    "model.layers.40.self_attn.q_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.40.self_attn.v_proj.bias": "model-00040-of-00046.safetensors",
+    "model.layers.40.self_attn.v_proj.weight": "model-00040-of-00046.safetensors",
+    "model.layers.41.input_layernorm.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.mlp.gate.e_score_correction_bias": "model-00042-of-00046.safetensors",
+    "model.layers.41.mlp.gate.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.mlp.moe_mlp.experts.weight": "model-00041-of-00046.safetensors",
+    "model.layers.41.mlp.moe_mlp.output_experts.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.mlp.shared_experts.down_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.mlp.shared_experts.gate_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.mlp.shared_experts.up_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.post_attention_layernorm.weight": "model-00042-of-00046.safetensors",
+    "model.layers.41.self_attn.k_proj.bias": "model-00041-of-00046.safetensors",
+    "model.layers.41.self_attn.k_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.41.self_attn.o_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.41.self_attn.q_proj.bias": "model-00041-of-00046.safetensors",
+    "model.layers.41.self_attn.q_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.41.self_attn.v_proj.bias": "model-00041-of-00046.safetensors",
+    "model.layers.41.self_attn.v_proj.weight": "model-00041-of-00046.safetensors",
+    "model.layers.42.input_layernorm.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.mlp.gate.e_score_correction_bias": "model-00043-of-00046.safetensors",
+    "model.layers.42.mlp.gate.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.mlp.moe_mlp.experts.weight": "model-00042-of-00046.safetensors",
+    "model.layers.42.mlp.moe_mlp.output_experts.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.mlp.shared_experts.down_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.mlp.shared_experts.gate_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.mlp.shared_experts.up_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.post_attention_layernorm.weight": "model-00043-of-00046.safetensors",
+    "model.layers.42.self_attn.k_proj.bias": "model-00042-of-00046.safetensors",
+    "model.layers.42.self_attn.k_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.42.self_attn.o_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.42.self_attn.q_proj.bias": "model-00042-of-00046.safetensors",
+    "model.layers.42.self_attn.q_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.42.self_attn.v_proj.bias": "model-00042-of-00046.safetensors",
+    "model.layers.42.self_attn.v_proj.weight": "model-00042-of-00046.safetensors",
+    "model.layers.43.input_layernorm.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.mlp.gate.e_score_correction_bias": "model-00044-of-00046.safetensors",
+    "model.layers.43.mlp.gate.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.mlp.moe_mlp.experts.weight": "model-00043-of-00046.safetensors",
+    "model.layers.43.mlp.moe_mlp.output_experts.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.mlp.shared_experts.down_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.mlp.shared_experts.gate_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.mlp.shared_experts.up_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.post_attention_layernorm.weight": "model-00044-of-00046.safetensors",
+    "model.layers.43.self_attn.k_proj.bias": "model-00043-of-00046.safetensors",
+    "model.layers.43.self_attn.k_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.43.self_attn.o_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.43.self_attn.q_proj.bias": "model-00043-of-00046.safetensors",
+    "model.layers.43.self_attn.q_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.43.self_attn.v_proj.bias": "model-00043-of-00046.safetensors",
+    "model.layers.43.self_attn.v_proj.weight": "model-00043-of-00046.safetensors",
+    "model.layers.44.input_layernorm.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.mlp.gate.e_score_correction_bias": "model-00045-of-00046.safetensors",
+    "model.layers.44.mlp.gate.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.mlp.moe_mlp.experts.weight": "model-00044-of-00046.safetensors",
+    "model.layers.44.mlp.moe_mlp.output_experts.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.mlp.shared_experts.down_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.mlp.shared_experts.gate_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.mlp.shared_experts.up_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.post_attention_layernorm.weight": "model-00045-of-00046.safetensors",
+    "model.layers.44.self_attn.k_proj.bias": "model-00044-of-00046.safetensors",
+    "model.layers.44.self_attn.k_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.44.self_attn.o_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.44.self_attn.q_proj.bias": "model-00044-of-00046.safetensors",
+    "model.layers.44.self_attn.q_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.44.self_attn.v_proj.bias": "model-00044-of-00046.safetensors",
+    "model.layers.44.self_attn.v_proj.weight": "model-00044-of-00046.safetensors",
+    "model.layers.45.input_layernorm.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.mlp.gate.e_score_correction_bias": "model-00046-of-00046.safetensors",
+    "model.layers.45.mlp.gate.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.mlp.moe_mlp.experts.weight": "model-00045-of-00046.safetensors",
+    "model.layers.45.mlp.moe_mlp.output_experts.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.mlp.shared_experts.down_proj.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.mlp.shared_experts.gate_proj.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.mlp.shared_experts.up_proj.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.post_attention_layernorm.weight": "model-00046-of-00046.safetensors",
+    "model.layers.45.self_attn.k_proj.bias": "model-00045-of-00046.safetensors",
+    "model.layers.45.self_attn.k_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.45.self_attn.o_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.45.self_attn.q_proj.bias": "model-00045-of-00046.safetensors",
+    "model.layers.45.self_attn.q_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.45.self_attn.v_proj.bias": "model-00045-of-00046.safetensors",
+    "model.layers.45.self_attn.v_proj.weight": "model-00045-of-00046.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.mlp.gate.e_score_correction_bias": "model-00006-of-00046.safetensors",
+    "model.layers.5.mlp.gate.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.mlp.moe_mlp.experts.weight": "model-00005-of-00046.safetensors",
+    "model.layers.5.mlp.moe_mlp.output_experts.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.mlp.shared_experts.down_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.mlp.shared_experts.gate_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.mlp.shared_experts.up_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00006-of-00046.safetensors",
+    "model.layers.5.self_attn.k_proj.bias": "model-00005-of-00046.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.5.self_attn.q_proj.bias": "model-00005-of-00046.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.5.self_attn.v_proj.bias": "model-00005-of-00046.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00005-of-00046.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.mlp.gate.e_score_correction_bias": "model-00007-of-00046.safetensors",
+    "model.layers.6.mlp.gate.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.mlp.moe_mlp.experts.weight": "model-00006-of-00046.safetensors",
+    "model.layers.6.mlp.moe_mlp.output_experts.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.mlp.shared_experts.down_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.mlp.shared_experts.gate_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.mlp.shared_experts.up_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00007-of-00046.safetensors",
+    "model.layers.6.self_attn.k_proj.bias": "model-00006-of-00046.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.6.self_attn.q_proj.bias": "model-00006-of-00046.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.6.self_attn.v_proj.bias": "model-00006-of-00046.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00006-of-00046.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.mlp.gate.e_score_correction_bias": "model-00008-of-00046.safetensors",
+    "model.layers.7.mlp.gate.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.mlp.moe_mlp.experts.weight": "model-00007-of-00046.safetensors",
+    "model.layers.7.mlp.moe_mlp.output_experts.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.mlp.shared_experts.down_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.mlp.shared_experts.gate_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.mlp.shared_experts.up_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00008-of-00046.safetensors",
+    "model.layers.7.self_attn.k_proj.bias": "model-00007-of-00046.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.7.self_attn.q_proj.bias": "model-00007-of-00046.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.7.self_attn.v_proj.bias": "model-00007-of-00046.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00007-of-00046.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.mlp.gate.e_score_correction_bias": "model-00009-of-00046.safetensors",
+    "model.layers.8.mlp.gate.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.mlp.moe_mlp.experts.weight": "model-00008-of-00046.safetensors",
+    "model.layers.8.mlp.moe_mlp.output_experts.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.mlp.shared_experts.down_proj.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.mlp.shared_experts.gate_proj.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.mlp.shared_experts.up_proj.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00009-of-00046.safetensors",
+    "model.layers.8.self_attn.k_proj.bias": "model-00008-of-00046.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.8.self_attn.q_proj.bias": "model-00008-of-00046.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.8.self_attn.v_proj.bias": "model-00008-of-00046.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00008-of-00046.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.mlp.gate.e_score_correction_bias": "model-00010-of-00046.safetensors",
+    "model.layers.9.mlp.gate.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.mlp.moe_mlp.experts.weight": "model-00009-of-00046.safetensors",
+    "model.layers.9.mlp.moe_mlp.output_experts.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.mlp.shared_experts.down_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.mlp.shared_experts.gate_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.mlp.shared_experts.up_proj.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00010-of-00046.safetensors",
+    "model.layers.9.self_attn.k_proj.bias": "model-00009-of-00046.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00009-of-00046.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00009-of-00046.safetensors",
+    "model.layers.9.self_attn.q_proj.bias": "model-00009-of-00046.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00009-of-00046.safetensors",
+    "model.layers.9.self_attn.v_proj.bias": "model-00009-of-00046.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00009-of-00046.safetensors",
+    "model.norm.weight": "model-00046-of-00046.safetensors"
+  }
+}
diff --git a/modeling_glm4_shared_moe.py b/modeling_glm4_shared_moe.py
new file mode 100644
index 0000000000000000000000000000000000000000..ca69415c8b96189226c56566f2acc5d5528fad22
--- /dev/null
+++ b/modeling_glm4_shared_moe.py
@@ -0,0 +1,642 @@
+# coding=utf-8
+# Copyright 2025 The ZhipuAI Inc. team and HuggingFace Inc. 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 Callable, Optional, Union
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import check_model_inputs
+from .configuration_glm4_shared_moe import Glm4SharedMoeConfig
+
+import scattermoe
+
+# Liger Patch #
+from liger_kernel.transformers.rms_norm import LigerRMSNorm as Glm4MoeRMSNorm
+from liger_kernel.transformers.model.loss_utils import LigerForCausalLMLoss
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Keep half or full tensor for later concatenation
+    rotary_dim = cos.shape[-1]
+    q_rot, q_pass = q[..., :rotary_dim], q[..., rotary_dim:]
+    k_rot, k_pass = k[..., :rotary_dim], k[..., rotary_dim:]
+
+    # Apply rotary embeddings on the first half or full tensor
+    q_embed = (q_rot * cos) + (rotate_half(q_rot) * sin)
+    k_embed = (k_rot * cos) + (rotate_half(k_rot) * sin)
+
+    # Concatenate back to full shape
+    q_embed = torch.cat([q_embed, q_pass], dim=-1)
+    k_embed = torch.cat([k_embed, k_pass], dim=-1)
+    return q_embed, k_embed
+
+
+class Glm4MoeAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: Glm4SharedMoeConfig, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.rope_scaling = config.rope_scaling
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+
+        self.q_proj = nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.k_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.v_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+        self.use_qk_norm = config.use_qk_norm
+        if self.use_qk_norm:
+            self.q_norm = Glm4MoeRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+            self.k_norm = Glm4MoeRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape)
+        key_states = self.k_proj(hidden_states).view(hidden_shape)
+        value_states = self.v_proj(hidden_states).view(hidden_shape)
+
+        if self.use_qk_norm:  # main diff from Llama
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_values is not None:
+            # sin and cos are specific to RoPE models; position_ids needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Glm4MoeMLP(nn.Module):
+    def __init__(self, config, hidden_size=None, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size if hidden_size is None else hidden_size
+        self.intermediate_size = config.intermediate_size if intermediate_size is None else intermediate_size
+
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+
+class Glm4MoeTopkRouter2(nn.Module):
+    def __init__(self, config: Glm4SharedMoeConfig):
+        super().__init__()
+        self.config = config
+        self.top_k = config.num_experts_per_tok
+        self.n_routed_experts = config.n_routed_experts
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.n_group = config.n_group
+        self.topk_group = config.topk_group
+        self.norm_topk_prob = config.norm_topk_prob
+
+        self.weight = nn.Parameter(torch.empty((self.n_routed_experts, config.hidden_size)))
+        self.register_buffer("e_score_correction_bias", torch.zeros((self.n_routed_experts), dtype=torch.float32))
+
+    def forward(self, hidden_states):
+        hidden_states = hidden_states.view(-1, self.config.hidden_size)
+        router_logits = F.linear(hidden_states.type(torch.float32), self.weight.type(torch.float32))
+        scores = router_logits.sigmoid() 
+        scores2 = scores + self.e_score_correction_bias.unsqueeze(0).to(scores.device)
+        _, selected_experts = torch.topk(
+            scores2, self.top_k, dim=-1
+        )
+        routing_weights = scores.gather(1, selected_experts)
+        if self.norm_topk_prob:
+            denominator = routing_weights.sum(dim=-1, keepdim=True) + 1e-20
+            routing_weights /= denominator
+        routing_weights = routing_weights * self.routed_scaling_factor
+        return selected_experts, routing_weights
+
+
+class Glm4MoeTopkRouter(nn.Module):
+    def __init__(self, config: Glm4SharedMoeConfig):
+        super().__init__()
+        self.config = config
+        self.top_k = config.num_experts_per_tok
+        self.n_routed_experts = config.n_routed_experts
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.n_group = config.n_group
+        self.topk_group = config.topk_group
+        self.norm_topk_prob = config.norm_topk_prob
+
+        self.weight = nn.Parameter(torch.empty((self.n_routed_experts, config.hidden_size)))
+        self.register_buffer("e_score_correction_bias", torch.zeros((self.n_routed_experts), dtype=torch.float32))
+
+    @torch.no_grad()
+    def get_topk_indices(self, scores):
+        scores_for_choice = scores.view(-1, self.n_routed_experts) + self.e_score_correction_bias.unsqueeze(0).to(scores.device)
+        group_scores = (
+            scores_for_choice.view(-1, self.n_group, self.n_routed_experts // self.n_group)
+            .topk(2, dim=-1)[0]
+            .sum(dim=-1)
+        )
+        group_idx = torch.topk(group_scores, k=self.topk_group, dim=-1, sorted=False)[1]
+        group_mask = torch.zeros_like(group_scores)
+        group_mask.scatter_(1, group_idx, 1)
+        score_mask = (
+            group_mask.unsqueeze(-1)
+            .expand(-1, self.n_group, self.n_routed_experts // self.n_group)
+            .reshape(-1, self.n_routed_experts)
+        )
+        scores_for_choice = scores_for_choice.masked_fill(~score_mask.bool(), 0.0)
+        topk_indices = torch.topk(scores_for_choice, k=self.top_k, dim=-1, sorted=False)[1]
+        return topk_indices
+
+    def forward(self, hidden_states):
+        hidden_states = hidden_states.view(-1, self.config.hidden_size)
+        router_logits = F.linear(hidden_states.type(torch.float32), self.weight.type(torch.float32))
+        scores = router_logits.sigmoid()
+        topk_indices = self.get_topk_indices(scores)
+        topk_weights = scores.gather(1, topk_indices)
+        if self.norm_topk_prob:
+            denominator = topk_weights.sum(dim=-1, keepdim=True) + 1e-20
+            topk_weights /= denominator
+        topk_weights = topk_weights * self.routed_scaling_factor
+        return topk_indices, topk_weights
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class Glm4MoeRMSNorm2(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        Glm4MoeRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+class Glm4MoeMoE(nn.Module):
+    """
+    A mixed expert module containing shared experts.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.moe_mlp = scattermoe.mlp.GLUMLP(
+            input_size=self.config.hidden_size,
+            hidden_size=self.config.moe_intermediate_size,
+            num_experts=self.config.n_routed_experts,
+            top_k=self.config.num_experts_per_tok,
+            activation=ACT2FN[config.hidden_act],
+        )
+        self.gate = Glm4MoeTopkRouter(config)
+        self.shared_experts = Glm4MoeMLP(
+            config=config, intermediate_size=config.moe_intermediate_size * config.n_shared_experts
+        )
+
+    def forward(self, hidden_states):
+        residuals = hidden_states
+        orig_shape = hidden_states.shape
+        topk_indices, topk_weights = self.gate(hidden_states)
+        hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+        topk_weights = topk_weights.to(torch.bfloat16)
+        hs_0 = self.moe_mlp(hidden_states, topk_weights, topk_indices).view(*orig_shape)
+        res = hs_0 + self.shared_experts(residuals)
+        return res
+
+
+class Glm4MoeDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: Glm4SharedMoeConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = Glm4MoeAttention(config=config, layer_idx=layer_idx)
+
+        if layer_idx >= config.first_k_dense_replace:
+            self.mlp = Glm4MoeMoE(config)
+        else:
+            self.mlp = Glm4MoeMLP(config)
+
+        self.input_layernorm = Glm4MoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = Glm4MoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,  # necessary, but kept here for BC
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+        # Self Attention
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+
+
+@auto_docstring
+class Glm4SharedMoePreTrainedModel(PreTrainedModel):
+    config: Glm4SharedMoeConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Glm4MoeDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _can_compile_fullgraph = False
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": Glm4MoeDecoderLayer,
+        "attentions": Glm4MoeAttention,
+    }
+
+    def _init_weights(self, module):
+        super()._init_weights(module)
+        if isinstance(module, Glm4MoeTopkRouter):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+
+
+class Glm4MoeRotaryEmbedding(nn.Module):
+    inv_freq: torch.Tensor  # fix linting for `register_buffer`
+
+    def __init__(self, config: Glm4SharedMoeConfig, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @torch.no_grad()
+    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):  # Force float32
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos() * self.attention_scaling
+            sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+@auto_docstring
+class Glm4SharedMoeModel(Glm4SharedMoePreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"model\.layers\.92.*", r"model\.layers\.46.*"]
+
+    def __init__(self, config: Glm4SharedMoeConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [Glm4MoeDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = Glm4MoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = Glm4MoeRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds: torch.Tensor = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache(config=self.config)
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position: torch.Tensor = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_values=past_key_values,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+@auto_docstring
+class Glm4SharedMoeForCausalLM(Glm4SharedMoePreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    _tp_plan = {"lm_head": "colwise_rep"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = Glm4SharedMoeModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> CausalLMOutputWithPast:
+        r"""
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, Glm4MoeForCausalLM
+
+        >>> model = Glm4MoeForCausalLM.from_pretrained("meta-glm4_moe/Glm4Moe-2-7b-hf")
+        >>> tokenizer = AutoTokenizer.from_pretrained("meta-glm4_moe/Glm4Moe-2-7b-hf")
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        outputs: BaseModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        kept_hidden_states = hidden_states[:, slice_indices, :]
+        shift_labels = kwargs.pop("shift_labels", None)
+        logits = None
+        loss = None
+
+        skip_logits = self.training and (labels is not None or shift_labels is not None)
+        if skip_logits:
+            loss = LigerForCausalLMLoss(
+                hidden_states=kept_hidden_states,
+                lm_head_weight=self.lm_head.weight,
+                labels=labels,
+                shift_labels=shift_labels,
+                hidden_size=self.config.hidden_size,
+                **kwargs,
+            )
+        else:
+            logits = self.lm_head(hidden_states[:, slice_indices, :])
+            if labels is not None:
+                loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+__all__ = ["Glm4SharedMoePreTrainedModel", "Glm4SharedMoeModel", "Glm4SharedMoeForCausalLM"]
diff --git a/modeling_glm4_shared_moe.py.bak b/modeling_glm4_shared_moe.py.bak
new file mode 100644
index 0000000000000000000000000000000000000000..3d817cf1d5ae8f99cffee80cd09a91dd7ff611cc
--- /dev/null
+++ b/modeling_glm4_shared_moe.py.bak
@@ -0,0 +1,628 @@
+# coding=utf-8
+# Copyright 2025 The ZhipuAI Inc. team and HuggingFace Inc. 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 Callable, Optional, Union
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import check_model_inputs
+from .configuration_glm4_shared_moe import Glm4SharedMoeConfig
+
+import scattermoe
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+
+    # Keep half or full tensor for later concatenation
+    rotary_dim = cos.shape[-1]
+    q_rot, q_pass = q[..., :rotary_dim], q[..., rotary_dim:]
+    k_rot, k_pass = k[..., :rotary_dim], k[..., rotary_dim:]
+
+    # Apply rotary embeddings on the first half or full tensor
+    q_embed = (q_rot * cos) + (rotate_half(q_rot) * sin)
+    k_embed = (k_rot * cos) + (rotate_half(k_rot) * sin)
+
+    # Concatenate back to full shape
+    q_embed = torch.cat([q_embed, q_pass], dim=-1)
+    k_embed = torch.cat([k_embed, k_pass], dim=-1)
+    return q_embed, k_embed
+
+
+class Glm4MoeAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: Glm4SharedMoeConfig, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.rope_scaling = config.rope_scaling
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+
+        self.q_proj = nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.k_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.v_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+        self.use_qk_norm = config.use_qk_norm
+        if self.use_qk_norm:
+            self.q_norm = Glm4MoeRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+            self.k_norm = Glm4MoeRMSNorm(self.head_dim, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape)
+        key_states = self.k_proj(hidden_states).view(hidden_shape)
+        value_states = self.v_proj(hidden_states).view(hidden_shape)
+
+        if self.use_qk_norm:  # main diff from Llama
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_values is not None:
+            # sin and cos are specific to RoPE models; position_ids needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Glm4MoeMLP(nn.Module):
+    def __init__(self, config, hidden_size=None, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size if hidden_size is None else hidden_size
+        self.intermediate_size = config.intermediate_size if intermediate_size is None else intermediate_size
+
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+
+class Glm4MoeTopkRouter(nn.Module):
+    def __init__(self, config: Glm4SharedMoeConfig):
+        super().__init__()
+        self.config = config
+        self.top_k = config.num_experts_per_tok
+        self.n_routed_experts = config.n_routed_experts
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.n_group = config.n_group
+        self.topk_group = config.topk_group
+        self.norm_topk_prob = config.norm_topk_prob
+
+        self.weight = nn.Parameter(torch.empty((self.n_routed_experts, config.hidden_size)))
+        self.register_buffer("e_score_correction_bias", torch.zeros((self.n_routed_experts), dtype=torch.float32))
+
+    @torch.no_grad()
+    def get_topk_indices(self, scores):
+        scores_for_choice = scores.view(-1, self.n_routed_experts) + self.e_score_correction_bias.unsqueeze(0).to(scores.device)
+        group_scores = (
+            scores_for_choice.view(-1, self.n_group, self.n_routed_experts // self.n_group)
+            .topk(2, dim=-1)[0]
+            .sum(dim=-1)
+        )
+        group_idx = torch.topk(group_scores, k=self.topk_group, dim=-1, sorted=False)[1]
+        group_mask = torch.zeros_like(group_scores)
+        group_mask.scatter_(1, group_idx, 1)
+        score_mask = (
+            group_mask.unsqueeze(-1)
+            .expand(-1, self.n_group, self.n_routed_experts // self.n_group)
+            .reshape(-1, self.n_routed_experts)
+        )
+        scores_for_choice = scores_for_choice.masked_fill(~score_mask.bool(), 0.0)
+        topk_indices = torch.topk(scores_for_choice, k=self.top_k, dim=-1, sorted=False)[1]
+        return topk_indices
+
+    def forward(self, hidden_states):
+        hidden_states = hidden_states.view(-1, self.config.hidden_size)
+        router_logits = F.linear(hidden_states.type(torch.float32), self.weight.type(torch.float32))
+        scores = router_logits.sigmoid()
+        topk_indices = self.get_topk_indices(scores)
+        topk_weights = scores.gather(1, topk_indices)
+        if self.norm_topk_prob:
+            denominator = topk_weights.sum(dim=-1, keepdim=True) + 1e-20
+            topk_weights /= denominator
+        topk_weights = topk_weights * self.routed_scaling_factor
+        return topk_indices, topk_weights
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class Glm4MoeRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        Glm4MoeRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+class Glm4MoeMoE(nn.Module):
+    """
+    A mixed expert module containing shared experts.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        # self.experts = nn.ModuleList(
+        #     [
+        #         Glm4MoeMLP(config, intermediate_size=config.moe_intermediate_size)
+        #         for _ in range(config.n_routed_experts)
+        #     ]
+        # )
+        self.moe_mlp = scattermoe.mlp.GLUMLP(
+            input_size=self.config.hidden_size,
+            hidden_size=self.config.moe_intermediate_size,
+            num_experts=self.config.n_routed_experts,
+            top_k=self.config.num_experts_per_tok,
+            activation=ACT2FN[config.hidden_act],
+        )
+        self.gate = Glm4MoeTopkRouter(config)
+        self.shared_experts = Glm4MoeMLP(
+            config=config, intermediate_size=config.moe_intermediate_size * config.n_shared_experts
+        )
+
+    #def moe(self, hidden_states: torch.Tensor, topk_indices: torch.Tensor, topk_weights: torch.Tensor):
+        # final_hidden_states = torch.zeros_like(hidden_states, dtype=topk_weights.dtype)
+        # expert_mask = torch.nn.functional.one_hot(topk_indices, num_classes=len(self.experts))
+        # expert_mask = expert_mask.permute(2, 0, 1)
+
+        # for expert_idx in range(len(self.experts)):
+        #     expert = self.experts[expert_idx]
+        #     mask = expert_mask[expert_idx]
+        #     token_indices, weight_indices = torch.where(mask)
+
+        #     if token_indices.numel() > 0:
+        #         expert_weights = topk_weights[token_indices, weight_indices]
+        #         expert_input = hidden_states[token_indices]
+        #         expert_output = expert(expert_input)
+        #         weighted_output = expert_output * expert_weights.unsqueeze(-1)
+        #         final_hidden_states.index_add_(0, token_indices, weighted_output)
+
+        #bsz, seq_len, h = hidden_states.shape
+        #hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+
+
+        # in original deepseek, the output of the experts are gathered once we leave this module
+        # thus the moe module is itelsf an IsolatedParallel module
+        # and all expert are "local" meaning we shard but we don't gather
+        #return final_hidden_states.type(hidden_states.dtype)
+
+    def forward(self, hidden_states):
+        # residuals = hidden_states
+        # orig_shape = hidden_states.shape
+        bsz, seq_len, h = hidden_states.shape
+        topk_indices, topk_weights = self.gate(hidden_states)
+        hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+        # hidden_states = self.moe(hidden_states, topk_indices, topk_weights).view(*orig_shape)
+        hidden_states = self.moe_mlp(hidden_states, topk_weights.to(torch.bfloat16), topk_indices)
+        hidden_states = hidden_states + self.shared_experts(hidden_states)
+        hidden_states = hidden_states.reshape(bsz, seq_len, h)
+        return hidden_states
+
+
+class Glm4MoeDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: Glm4SharedMoeConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = Glm4MoeAttention(config=config, layer_idx=layer_idx)
+
+        if layer_idx >= config.first_k_dense_replace:
+            self.mlp = Glm4MoeMoE(config)
+        else:
+            self.mlp = Glm4MoeMLP(config)
+
+        self.input_layernorm = Glm4MoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = Glm4MoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,  # necessary, but kept here for BC
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+        # Self Attention
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+
+
+@auto_docstring
+class Glm4SharedMoePreTrainedModel(PreTrainedModel):
+    config: Glm4SharedMoeConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Glm4MoeDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _can_compile_fullgraph = False
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": Glm4MoeDecoderLayer,
+        "attentions": Glm4MoeAttention,
+    }
+
+    def _init_weights(self, module):
+        super()._init_weights(module)
+        if isinstance(module, Glm4MoeTopkRouter):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+
+
+class Glm4MoeRotaryEmbedding(nn.Module):
+    inv_freq: torch.Tensor  # fix linting for `register_buffer`
+
+    def __init__(self, config: Glm4SharedMoeConfig, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @torch.no_grad()
+    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):  # Force float32
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos() * self.attention_scaling
+            sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+@auto_docstring
+class Glm4SharedMoeModel(Glm4SharedMoePreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"model\.layers\.92.*", r"model\.layers\.46.*"]
+
+    def __init__(self, config: Glm4SharedMoeConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [Glm4MoeDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = Glm4MoeRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = Glm4MoeRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds: torch.Tensor = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache(config=self.config)
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position: torch.Tensor = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_values=past_key_values,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+@auto_docstring
+class Glm4SharedMoeForCausalLM(Glm4SharedMoePreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    _tp_plan = {"lm_head": "colwise_rep"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = Glm4SharedMoeModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> CausalLMOutputWithPast:
+        r"""
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, Glm4MoeForCausalLM
+
+        >>> model = Glm4MoeForCausalLM.from_pretrained("meta-glm4_moe/Glm4Moe-2-7b-hf")
+        >>> tokenizer = AutoTokenizer.from_pretrained("meta-glm4_moe/Glm4Moe-2-7b-hf")
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        outputs: BaseModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+__all__ = ["Glm4SharedMoePreTrainedModel", "Glm4SharedMoeModel", "Glm4SharedMoeForCausalLM"]
diff --git a/tokenizer.json b/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..43a71efad1ab3f7e4d66d1e48e7cc44f68f21f3a
--- /dev/null
+++ b/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9340665016419c825c4bdabbcc9acc43b7ca2c68ce142724afa829abb1be5efd
+size 19970699
diff --git a/tokenizer_config.json b/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..053f12c9b03d5acbcc921042ea0c87a6baa5d3f7
--- /dev/null
+++ b/tokenizer_config.json
@@ -0,0 +1,325 @@
+{
+  "added_tokens_decoder": {
+    "151329": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151330": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151331": {
+      "content": "[gMASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151332": {
+      "content": "[sMASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151333": {
+      "content": "<sop>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151334": {
+      "content": "<eop>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151335": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151336": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151337": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151338": {
+      "content": "<|observation|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151339": {
+      "content": "<|begin_of_image|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151340": {
+      "content": "<|end_of_image|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151341": {
+      "content": "<|begin_of_video|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151342": {
+      "content": "<|end_of_video|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151343": {
+      "content": "<|begin_of_audio|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151344": {
+      "content": "<|end_of_audio|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151345": {
+      "content": "<|begin_of_transcription|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151346": {
+      "content": "<|end_of_transcription|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151347": {
+      "content": "<|code_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151348": {
+      "content": "<|code_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151349": {
+      "content": "<|code_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151350": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151351": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151352": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151353": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151354": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151355": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151356": {
+      "content": "<arg_key>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151357": {
+      "content": "</arg_key>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151358": {
+      "content": "<arg_value>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151359": {
+      "content": "</arg_value>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151360": {
+      "content": "/nothink",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151361": {
+      "content": "<|begin_of_box|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151362": {
+      "content": "<|end_of_box|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151363": {
+      "content": "<|image|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151364": {
+      "content": "<|video|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|endoftext|>",
+    "[MASK]",
+    "[gMASK]",
+    "[sMASK]",
+    "<sop>",
+    "<eop>",
+    "<|system|>",
+    "<|user|>",
+    "<|assistant|>",
+    "<|observation|>",
+    "<|begin_of_image|>",
+    "<|end_of_image|>",
+    "<|begin_of_video|>",
+    "<|end_of_video|>",
+    "<|begin_of_audio|>",
+    "<|end_of_audio|>",
+    "<|begin_of_transcription|>",
+    "<|end_of_transcription|>",
+    "<|code_prefix|>",
+    "<|code_middle|>",
+    "<|code_suffix|>",
+    "/nothink"
+  ],
+  "clean_up_tokenization_spaces": false,
+  "do_lower_case": false,
+  "eos_token": "<|endoftext|>",
+  "extra_special_tokens": {},
+  "model_max_length": 128000,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "left",
+  "remove_space": false,
+  "tokenizer_class": "PreTrainedTokenizer"
+}