Upload folder using huggingface_hub

config.json

@@ -0,0 +1,42 @@
+{
+  "architectures": [
+    "Qwen3SharedMoeForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_qwen3_shared_moe.Qwen3SharedMoeConfig",
+    "AutoModel": "modeling_qwen3_shared_moe.Qwen3SharedMoeModel",
+    "AutoModelForCausalLM": "modeling_qwen3_shared_moe.Qwen3SharedMoeForCausalLM"
+  },
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "decoder_sparse_step": 1,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5472,
+  "max_position_embeddings": 262144,
+  "max_window_layers": 28,
+  "mlp_only_layers": [],
+  "model_type": "qwen3_shared_moe",
+  "moe_intermediate_size": 768,
+  "norm_topk_prob": true,
+  "num_attention_heads": 32,
+  "num_experts": 128,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 48,
+  "num_key_value_heads": 4,
+  "output_router_logits": false,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000000,
+  "router_aux_loss_coef": 0.0,
+  "shared_expert_intermediate_size": null,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.54.1",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

configuration_qwen3_shared_moe.py

@@ -0,0 +1,230 @@
+# coding=utf-8
+# Copyright 2024 The Qwen team, Alibaba Group and the 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.
+"""Qwen3SharedMoE model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class Qwen3SharedMoeConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`Qwen3SharedMoeModel`]. It is used to instantiate a
+    Qwen3SharedMoE model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of [Qwen/Qwen3-MoE-15B-A2B](https://huggingface.co/Qwen/Qwen3-15B-A2B).
+
+    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 151936):
+            Vocabulary size of the Qwen3SharedMoE model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`Qwen3SharedMoeModel`]
+        hidden_size (`int`, *optional*, defaults to 2048):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 6144):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 24):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 4):
+            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 checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). 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 32768):
+            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-06):
+            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.
+        use_sliding_window (`bool`, *optional*, defaults to `False`):
+            Whether to use sliding window attention.
+        sliding_window (`int`, *optional*, defaults to 4096):
+            Sliding window attention (SWA) window size. If not specified, will default to `4096`.
+        max_window_layers (`int`, *optional*, defaults to 28):
+            The number of layers that use SWA (Sliding Window Attention). The bottom layers use SWA while the top use full attention.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        decoder_sparse_step (`int`, *optional*, defaults to 1):
+            The frequency of the MoE layer.
+        moe_intermediate_size (`int`, *optional*, defaults to 768):
+            Intermediate size of the routed expert.
+        shared_expert_intermediate_size (`int`, *optional*, defaults to None):
+            Intermediate size of the shared expert. `None` means no shared expert.
+        num_experts_per_tok (`int`, *optional*, defaults to 8):
+            Number of selected experts.
+        num_experts (`int`, *optional*, defaults to 128):
+            Number of routed experts.
+        norm_topk_prob (`bool`, *optional*, defaults to `False`):
+            Whether to normalize the topk probabilities.
+        output_router_logits (`bool`, *optional*, defaults to `False`):
+            Whether or not the router logits should be returned by the model. Enabling this will also
+            allow the model to output the auxiliary loss, including load balancing loss and router z-loss.
+        router_aux_loss_coef (`float`, *optional*, defaults to 0.001):
+            The aux loss factor for the total loss.
+        mlp_only_layers (`List[int]`, *optional*, defaults to `[]`):
+            Indicate which layers use Qwen3SharedMoeMLP rather than Qwen3SharedMoeSparseMoeBlock
+            The list contains layer index, from 0 to num_layers-1 if we have num_layers layers
+            If `mlp_only_layers` is empty, `decoder_sparse_step` is used to determine the sparsity.
+    """
+
+    model_type = "qwen3_shared_moe"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    # Default tensor parallel plan for base model `Qwen3SharedMoe`
+    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.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=151936,
+        hidden_size=2048,
+        intermediate_size=6144,
+        num_hidden_layers=24,
+        num_attention_heads=32,
+        num_key_value_heads=4,
+        hidden_act="silu",
+        max_position_embeddings=32768,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        use_sliding_window=False,
+        sliding_window=4096,
+        max_window_layers=28,
+        attention_dropout=0.0,
+        decoder_sparse_step=1,
+        moe_intermediate_size=768,
+        shared_expert_intermediate_size=None,
+        num_experts_per_tok=8,
+        num_experts=128,
+        norm_topk_prob=False,
+        output_router_logits=False,
+        router_aux_loss_coef=0.001,
+        mlp_only_layers=None,
+        **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.use_sliding_window = use_sliding_window
+        self.sliding_window = sliding_window if use_sliding_window else None
+        self.max_window_layers = max_window_layers
+
+        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.decoder_sparse_step = decoder_sparse_step
+        self.moe_intermediate_size = moe_intermediate_size
+        self.shared_expert_intermediate_size = shared_expert_intermediate_size
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_experts = num_experts
+        self.norm_topk_prob = norm_topk_prob
+        self.output_router_logits = output_router_logits
+        self.router_aux_loss_coef = router_aux_loss_coef
+        self.mlp_only_layers = [] if mlp_only_layers is None else mlp_only_layers
+
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+

generation_config.json

@@ -0,0 +1,4 @@
+{
+  "_from_model_config": true,
+  "transformers_version": "4.54.1"
+}

model-00001-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e4bc51223729e7e94e46819a7ad47bc605e0004ceb2afdc1f1cc158b6f7412a
+size 4399332008

model-00002-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0abfac116abb48dafc3989c05a006613aa159288391987c5afc7ed552d1c0358
+size 4984970264

model-00003-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9e0839d2e34a7f9fc2f81460d838a4c9273a5be6b9fff0551feee6c802f273a
+size 4984970272

model-00004-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fa85af58064c6226847717dad61f64ec7ba325710140a9f916f5843e40451d8
+size 4984970312

model-00005-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3f6e9fe1e4366945748ea4129b243b97d702dfa7cf560a9383bd283d1f6a4c9
+size 4984970312

model-00006-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5dccf2dbcb28d881f9b80904c8948430f54bc6d47d6825e7312eb1ae6027bdf5
+size 4984970312

model-00007-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edcb72301e7d2dbd6a8512fdf364e17f527fa91707feb72a83726772cdf1316c
+size 4984970312

model-00008-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0334caf6b8e9124e85801fd6a41961e4aa892342f0b2bbba065e1e167b0c2567
+size 4984970312

model-00009-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9429fb9ea9d57f1a5ec32dfa49e0434b8cba6bb7855f7c20a681503e559d78a2
+size 4984970312

model-00010-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:755719f4dd11111c0037dc923043af8338ee3d8b2c045dd6fae87ae0c0dc7658
+size 4984970312

model-00011-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c417dd8b3e5168b62fccfd70dc1be6d7f6c8743104d58805b24e02ad544af90
+size 4984970312

model-00012-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60496189195377cd81eb74843c0f184ad6f0ebd0a6cc34a02f7fdfcabdbc1a29
+size 4984970312

model-00013-of-00013.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ea4a7adc5c961c45d3273bd52e71c8bed2a54d70ff18596e066bfd77c46cb92
+size 1830302400

model.safetensors.index.json

@@ -0,0 +1,539 @@
+{
+  "metadata": {
+    "total_parameters": 30532122624,
+    "total_size": 61064245248
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00013-of-00013.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.mlp.gate.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.mlp.moe_mlp.experts.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.mlp.moe_mlp.output_experts.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.self_attn.k_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.self_attn.q_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.mlp.gate.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.mlp.moe_mlp.experts.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.mlp.moe_mlp.output_experts.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.self_attn.k_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.self_attn.q_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.mlp.gate.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.mlp.moe_mlp.experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.mlp.moe_mlp.output_experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.self_attn.k_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.self_attn.q_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.11.mlp.gate.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.mlp.moe_mlp.experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.11.mlp.moe_mlp.output_experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.11.self_attn.k_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.self_attn.q_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.mlp.gate.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.mlp.moe_mlp.experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.mlp.moe_mlp.output_experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.self_attn.k_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.self_attn.q_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.mlp.gate.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.mlp.moe_mlp.experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.mlp.moe_mlp.output_experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.self_attn.k_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.self_attn.q_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.mlp.gate.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.mlp.moe_mlp.experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.mlp.moe_mlp.output_experts.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.self_attn.k_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.self_attn.q_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.15.mlp.gate.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.mlp.moe_mlp.experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.15.mlp.moe_mlp.output_experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.15.self_attn.k_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.self_attn.q_norm.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00004-of-00013.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.mlp.gate.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.mlp.moe_mlp.experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.mlp.moe_mlp.output_experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.self_attn.k_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.self_attn.q_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.mlp.gate.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.mlp.moe_mlp.experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.mlp.moe_mlp.output_experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.self_attn.k_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.self_attn.q_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.mlp.gate.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.mlp.moe_mlp.experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.mlp.moe_mlp.output_experts.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.self_attn.k_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.self_attn.q_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.19.mlp.gate.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.mlp.moe_mlp.experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.19.mlp.moe_mlp.output_experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.19.self_attn.k_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.self_attn.q_norm.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00005-of-00013.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.mlp.gate.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.mlp.moe_mlp.experts.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.mlp.moe_mlp.output_experts.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.self_attn.k_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.self_attn.q_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.mlp.gate.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.mlp.moe_mlp.experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.mlp.moe_mlp.output_experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.self_attn.k_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.self_attn.q_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.mlp.gate.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.mlp.moe_mlp.experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.mlp.moe_mlp.output_experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.self_attn.k_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.self_attn.q_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.mlp.gate.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.mlp.moe_mlp.experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.mlp.moe_mlp.output_experts.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.self_attn.k_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.self_attn.q_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.23.mlp.gate.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.mlp.moe_mlp.experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.23.mlp.moe_mlp.output_experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.23.self_attn.k_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.self_attn.q_norm.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00006-of-00013.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.mlp.gate.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.mlp.moe_mlp.experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.mlp.moe_mlp.output_experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.self_attn.k_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.self_attn.q_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.mlp.gate.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.mlp.moe_mlp.experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.mlp.moe_mlp.output_experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.self_attn.k_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.self_attn.q_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.mlp.gate.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.mlp.moe_mlp.experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.mlp.moe_mlp.output_experts.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.self_attn.k_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.self_attn.q_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.27.mlp.gate.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.mlp.moe_mlp.experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.27.mlp.moe_mlp.output_experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.27.self_attn.k_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.self_attn.q_norm.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00007-of-00013.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.mlp.gate.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.mlp.moe_mlp.experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.mlp.moe_mlp.output_experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.self_attn.k_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.self_attn.q_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.mlp.gate.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.mlp.moe_mlp.experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.mlp.moe_mlp.output_experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.self_attn.k_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.self_attn.q_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.3.mlp.gate.weight": "model-00001-of-00013.safetensors",
+    "model.layers.3.mlp.moe_mlp.experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.3.mlp.moe_mlp.output_experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.3.self_attn.k_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.3.self_attn.q_norm.weight": "model-00001-of-00013.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00013.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.mlp.gate.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.mlp.moe_mlp.experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.mlp.moe_mlp.output_experts.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.self_attn.k_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.self_attn.q_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.31.mlp.gate.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.mlp.moe_mlp.experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.31.mlp.moe_mlp.output_experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.31.self_attn.k_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.self_attn.q_norm.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00008-of-00013.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.mlp.gate.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.mlp.moe_mlp.experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.mlp.moe_mlp.output_experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.self_attn.k_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.self_attn.q_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.mlp.gate.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.mlp.moe_mlp.experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.mlp.moe_mlp.output_experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.self_attn.k_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.self_attn.q_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.mlp.gate.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.mlp.moe_mlp.experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.mlp.moe_mlp.output_experts.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.self_attn.k_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.self_attn.q_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.35.mlp.gate.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.mlp.moe_mlp.experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.35.mlp.moe_mlp.output_experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.35.self_attn.k_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.self_attn.q_norm.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00009-of-00013.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.mlp.gate.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.mlp.moe_mlp.experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.mlp.moe_mlp.output_experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.self_attn.k_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.self_attn.q_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.mlp.gate.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.mlp.moe_mlp.experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.mlp.moe_mlp.output_experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.self_attn.k_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.self_attn.q_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.mlp.gate.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.mlp.moe_mlp.experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.mlp.moe_mlp.output_experts.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.self_attn.k_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.self_attn.q_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.39.mlp.gate.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.mlp.moe_mlp.experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.39.mlp.moe_mlp.output_experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.39.self_attn.k_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.self_attn.q_norm.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00010-of-00013.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.mlp.gate.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.mlp.moe_mlp.experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.mlp.moe_mlp.output_experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.self_attn.k_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.self_attn.q_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.40.input_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.mlp.gate.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.mlp.moe_mlp.experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.mlp.moe_mlp.output_experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.post_attention_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.self_attn.k_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.self_attn.k_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.self_attn.o_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.self_attn.q_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.self_attn.q_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.40.self_attn.v_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.input_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.mlp.gate.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.mlp.moe_mlp.experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.mlp.moe_mlp.output_experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.post_attention_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.self_attn.k_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.self_attn.k_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.self_attn.o_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.self_attn.q_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.self_attn.q_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.41.self_attn.v_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.input_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.mlp.gate.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.mlp.moe_mlp.experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.mlp.moe_mlp.output_experts.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.post_attention_layernorm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.self_attn.k_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.self_attn.k_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.self_attn.o_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.self_attn.q_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.self_attn.q_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.42.self_attn.v_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.input_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.43.mlp.gate.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.mlp.moe_mlp.experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.43.mlp.moe_mlp.output_experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.43.post_attention_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.43.self_attn.k_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.self_attn.k_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.self_attn.o_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.self_attn.q_norm.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.self_attn.q_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.43.self_attn.v_proj.weight": "model-00011-of-00013.safetensors",
+    "model.layers.44.input_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.mlp.gate.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.mlp.moe_mlp.experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.mlp.moe_mlp.output_experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.post_attention_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.self_attn.k_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.self_attn.k_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.self_attn.o_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.self_attn.q_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.self_attn.q_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.44.self_attn.v_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.input_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.mlp.gate.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.mlp.moe_mlp.experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.mlp.moe_mlp.output_experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.post_attention_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.self_attn.k_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.self_attn.k_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.self_attn.o_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.self_attn.q_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.self_attn.q_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.45.self_attn.v_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.input_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.mlp.gate.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.mlp.moe_mlp.experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.mlp.moe_mlp.output_experts.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.post_attention_layernorm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.self_attn.k_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.self_attn.k_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.self_attn.o_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.self_attn.q_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.self_attn.q_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.46.self_attn.v_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.input_layernorm.weight": "model-00013-of-00013.safetensors",
+    "model.layers.47.mlp.gate.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.mlp.moe_mlp.experts.weight": "model-00013-of-00013.safetensors",
+    "model.layers.47.mlp.moe_mlp.output_experts.weight": "model-00013-of-00013.safetensors",
+    "model.layers.47.post_attention_layernorm.weight": "model-00013-of-00013.safetensors",
+    "model.layers.47.self_attn.k_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.self_attn.k_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.self_attn.o_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.self_attn.q_norm.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.self_attn.q_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.47.self_attn.v_proj.weight": "model-00012-of-00013.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.mlp.gate.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.mlp.moe_mlp.experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.mlp.moe_mlp.output_experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.self_attn.k_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.self_attn.q_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.mlp.gate.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.mlp.moe_mlp.experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.mlp.moe_mlp.output_experts.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.self_attn.k_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.self_attn.q_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.7.mlp.gate.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.mlp.moe_mlp.experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.7.mlp.moe_mlp.output_experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.7.self_attn.k_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.self_attn.q_norm.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00002-of-00013.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.mlp.gate.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.mlp.moe_mlp.experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.mlp.moe_mlp.output_experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.self_attn.k_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.self_attn.q_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.mlp.gate.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.mlp.moe_mlp.experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.mlp.moe_mlp.output_experts.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.self_attn.k_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.self_attn.q_norm.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00003-of-00013.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00003-of-00013.safetensors",
+    "model.norm.weight": "model-00013-of-00013.safetensors"
+  }
+}

modeling_qwen3_shared_moe.py

@@ -0,0 +1,253 @@
+# coding=utf-8
+# Copyright 2025 Charles O. Goddard, The Qwen team, Alibaba Group and the 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.
+"""PyTorch Qwen3 model with shared expert support."""
+
+from typing import List, Optional, Union
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from transformers.modeling_outputs import (
+    MoeCausalLMOutputWithPast,
+    MoeModelOutputWithPast,
+)
+from transformers.activations import ACT2FN
+from transformers.utils import logging
+from transformers.models.mixtral.modeling_mixtral import (
+    load_balancing_loss_func,
+)
+from transformers.models.qwen3_moe.modeling_qwen3_moe import (
+    Qwen3MoeMLP,
+    Qwen3MoeRMSNorm,
+    Qwen3MoeAttention,
+    Qwen3MoeDecoderLayer,
+    Qwen3MoeModel,
+    Qwen3MoeForCausalLM,
+)
+from .configuration_qwen3_shared_moe import Qwen3SharedMoeConfig
+
+import scattermoe
+
+
+logger = logging.get_logger(__name__)
+
+
+class Qwen3SharedMoeSparseMoeBlock(nn.Module):
+    def __init__(self, config: Qwen3SharedMoeConfig):
+        super().__init__()
+        self.config = config
+        self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False)
+        if config.shared_expert_intermediate_size is not None:
+            self.shared_expert = Qwen3MoeMLP(
+                config, intermediate_size=config.shared_expert_intermediate_size
+            )
+        else:
+            self.shared_expert = None
+        self.moe_mlp = scattermoe.mlp.GLUMLP(
+            input_size=self.config.hidden_size,
+            hidden_size=self.config.moe_intermediate_size,
+            num_experts=self.config.num_experts,
+            top_k=self.config.num_experts_per_tok,
+            activation=ACT2FN[config.hidden_act],
+        )
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        # handling of gate/router logits copied from Qwen3MoeSparseMoeBlock
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+        # router_logits: (batch * sequence_length, n_experts)
+        router_logits = self.gate(hidden_states)
+
+        routing_weights = F.softmax(router_logits, dim=1, dtype=torch.float)
+        routing_weights, selected_experts = torch.topk(
+            routing_weights, self.config.num_experts_per_tok, dim=-1
+        )
+        if self.config.norm_topk_prob:  # only diff with mixtral sparse moe block!
+            routing_weights /= routing_weights.sum(dim=-1, keepdim=True)
+        # we cast back to the input dtype
+        routing_weights = routing_weights.to(hidden_states.dtype)
+
+        # modified here to use scattermoe + shared_expert
+        hs_0 = self.moe_mlp(hidden_states, routing_weights, selected_experts)
+
+        if self.shared_expert is not None:
+            shared_res = self.shared_expert(hidden_states)
+            res = hs_0 + shared_res
+        else:
+            res = hs_0
+        res = res.reshape(batch_size, sequence_length, hidden_dim)
+        return res, router_logits
+
+
+class Qwen3SharedMoeDecoderLayer(Qwen3MoeDecoderLayer, nn.Module):
+    def __init__(self, config: Qwen3SharedMoeConfig, layer_idx: int):
+        super().__init__(config, layer_idx)
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = Qwen3MoeAttention(config, layer_idx)
+
+        if (layer_idx not in config.mlp_only_layers) and (
+            config.num_experts > 0 and (layer_idx + 1) % config.decoder_sparse_step == 0
+        ):
+            self.mlp = Qwen3SharedMoeSparseMoeBlock(config)
+        else:
+            self.mlp = Qwen3MoeMLP(config, intermediate_size=config.intermediate_size)
+
+        self.input_layernorm = Qwen3MoeRMSNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
+        self.post_attention_layernorm = Qwen3MoeRMSNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
+
+
+class Qwen3SharedMoeModel(Qwen3MoeModel):
+    config_class = Qwen3SharedMoeConfig
+
+    def __init__(self, config: Qwen3SharedMoeConfig):
+        super().__init__(config)
+        self.layers = nn.ModuleList(
+            [
+                Qwen3SharedMoeDecoderLayer(config, layer_idx)
+                for layer_idx in range(config.num_hidden_layers)
+            ]
+        )
+
+
+class Qwen3SharedMoeForCausalLM(Qwen3MoeForCausalLM):
+    config_class = Qwen3SharedMoeConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = Qwen3SharedMoeModel(config)
+        self.num_experts = config.num_experts
+
+    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[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs,
+    ) -> MoeCausalLMOutputWithPast:
+        r"""
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+            logits_to_keep (`int` or `torch.Tensor`, *optional*):
+                If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+                `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+                token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+                If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+                This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, Qwen3MoeForCausalLM
+
+        >>> model = Qwen3MoeForCausalLM.from_pretrained("Qwen/Qwen3-MoE-15B-A2B")
+        >>> tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-MoE-15B-A2B")
+
+        >>> 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."
+        ```"""
+
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_router_logits = (
+            output_router_logits
+            if output_router_logits is not None
+            else self.config.output_router_logits
+        )
+
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs: MoeModelOutputWithPast = 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,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            output_router_logits=output_router_logits,
+            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, labels, self.vocab_size, **kwargs)
+
+        aux_loss = None
+        if output_router_logits:
+            aux_loss = load_balancing_loss_func(
+                outputs.router_logits,
+                self.num_experts,
+                self.num_experts_per_tok,
+                attention_mask,
+            )
+            if labels is not None:
+                loss += self.router_aux_loss_coef * aux_loss.to(
+                    loss.device
+                )  # make sure to reside in the same device
+
+        return MoeCausalLMOutputWithPast(
+            loss=loss,
+            aux_loss=aux_loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_logits=outputs.router_logits,
+        )
+

tokenizer_config.json

@@ -0,0 +1,239 @@
+{
+    "add_prefix_space": false,
+    "added_tokens_decoder": {
+        "151643": {
+            "content": "<|endoftext|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151644": {
+            "content": "<|im_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151645": {
+            "content": "<|im_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151646": {
+            "content": "<|object_ref_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151647": {
+            "content": "<|object_ref_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151648": {
+            "content": "<|box_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151649": {
+            "content": "<|box_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151650": {
+            "content": "<|quad_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151651": {
+            "content": "<|quad_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151652": {
+            "content": "<|vision_start|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151653": {
+            "content": "<|vision_end|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151654": {
+            "content": "<|vision_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151655": {
+            "content": "<|image_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151656": {
+            "content": "<|video_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": true
+        },
+        "151657": {
+            "content": "<tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151658": {
+            "content": "</tool_call>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151659": {
+            "content": "<|fim_prefix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151660": {
+            "content": "<|fim_middle|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151661": {
+            "content": "<|fim_suffix|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151662": {
+            "content": "<|fim_pad|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151663": {
+            "content": "<|repo_name|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151664": {
+            "content": "<|file_sep|>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151665": {
+            "content": "<tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151666": {
+            "content": "</tool_response>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151667": {
+            "content": "<think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        },
+        "151668": {
+            "content": "</think>",
+            "lstrip": false,
+            "normalized": false,
+            "rstrip": false,
+            "single_word": false,
+            "special": false
+        }
+    },
+    "additional_special_tokens": [
+        "<|im_start|>",
+        "<|im_end|>",
+        "<|object_ref_start|>",
+        "<|object_ref_end|>",
+        "<|box_start|>",
+        "<|box_end|>",
+        "<|quad_start|>",
+        "<|quad_end|>",
+        "<|vision_start|>",
+        "<|vision_end|>",
+        "<|vision_pad|>",
+        "<|image_pad|>",
+        "<|video_pad|>"
+    ],
+    "bos_token": null,
+    "chat_template": "{% macro render_item_list(item_list, tag_name='required') %}\n    {%- if item_list is defined and item_list is iterable and item_list | length > 0 %}\n        {%- if tag_name %}{{- '\\n<' ~ tag_name ~ '>' -}}{% endif %}\n            {{- '[' }}\n                {%- for item in item_list -%}\n                    {%- if loop.index > 1 %}{{- \", \"}}{% endif -%}\n                    {%- if item is string -%}\n                        {{ \"`\" ~ item ~ \"`\" }}\n                    {%- else -%}\n                        {{ item }}\n                    {%- endif -%}\n                {%- endfor -%}\n            {{- ']' }}\n        {%- if tag_name %}{{- '</' ~ tag_name ~ '>' -}}{% endif %}\n    {%- endif %}\n{% endmacro %}\n\n{%- if messages[0][\"role\"] == \"system\" %}\n    {%- set system_message = messages[0][\"content\"] %}\n    {%- set loop_messages = messages[1:] %}\n{%- else %}\n    {%- set loop_messages = messages %}\n{%- endif %}\n\n{%- if not tools is defined %}\n    {%- set tools = [] %}\n{%- endif %}\n\n{%- if system_message is defined %}\n    {{- \"<|im_start|>system\\n\" + system_message }}\n{%- else %}\n    {%- if tools is iterable and tools | length > 0 %}\n        {{- \"<|im_start|>system\\nYou are Qwen, a helpful AI assistant that can interact with a computer to solve tasks.\" }}\n    {%- endif %}\n{%- endif %}\n{%- if tools is iterable and tools | length > 0 %}\n    {{- \"\\n\\nYou have access to the following functions:\\n\\n\" }}\n    {{- \"<tools>\" }}\n    {%- for tool in tools %}\n        {%- if tool.function is defined %}\n            {%- set tool = tool.function %}\n        {%- endif %}\n        {{- \"\\n<function>\\n<name>\" ~ tool.name ~ \"</name>\" }}\n        {{- '\\n<description>' ~ (tool.description | trim) ~ '</description>' }}\n        {{- '\\n<parameters>' }}\n        {%- for param_name, param_fields in tool.parameters.properties|items %}\n            {{- '\\n<parameter>' }}\n            {{- '\\n<name>' ~ param_name ~ '</name>' }}\n            {%- if param_fields.type is defined %}\n                {{- '\\n<type>' ~ (param_fields.type | string) ~ '</type>' }}\n            {%- endif %}\n            {%- if param_fields.description is defined %}\n                {{- '\\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}\n            {%- endif %}\n            {{- render_item_list(param_fields.enum, 'enum') }}\n            {%- set handled_keys = ['type', 'description', 'enum', 'required'] %}\n            {%- for json_key in param_fields.keys() | reject(\"in\", handled_keys) %}\n                {%- set normed_json_key = json_key | replace(\"-\", \"_\") | replace(\" \", \"_\") | replace(\"$\", \"\") %}\n                {%- if param_fields[json_key] is mapping %}\n                    {{- '\\n<' ~ normed_json_key ~ '>' ~ (param_fields[json_key] | tojson | safe) ~ '</' ~ normed_json_key ~ '>' }}\n                {%- else %}\n                    {{-'\\n<' ~ normed_json_key ~ '>' ~ (param_fields[json_key] | string) ~ '</' ~ normed_json_key ~ '>' }}\n                {%- endif %}\n            {%- endfor %}\n            {{- render_item_list(param_fields.required, 'required') }}\n            {{- '\\n</parameter>' }}\n        {%- endfor %}\n        {{- render_item_list(tool.parameters.required, 'required') }}\n        {{- '\\n</parameters>' }}\n        {%- if tool.return is defined %}\n            {%- if tool.return is mapping %}\n                {{- '\\n<return>' ~ (tool.return | tojson | safe) ~ '</return>' }}\n            {%- else %}\n                {{- '\\n<return>' ~ (tool.return | string) ~ '</return>' }}\n            {%- endif %}\n        {%- endif %}\n        {{- '\\n</function>' }}\n    {%- endfor %}\n    {{- \"\\n</tools>\" }}\n    {{- '\\n\\nIf you choose to call a function ONLY reply in the following format with NO suffix:\\n\\n<tool_call>\\n<function=example_function_name>\\n<parameter=example_parameter_1>\\nvalue_1\\n</parameter>\\n<parameter=example_parameter_2>\\nThis is the value for the second parameter\\nthat can span\\nmultiple lines\\n</parameter>\\n</function>\\n</tool_call>\\n\\n<IMPORTANT>\\nReminder:\\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\\n- Required parameters MUST be specified\\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\\n</IMPORTANT>' }}\n{%- endif %}\n{%- if system_message is defined %}\n    {{- '<|im_end|>\\n' }}\n{%- else %}\n    {%- if tools is iterable and tools | length > 0 %}\n        {{- '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in loop_messages %}\n    {%- if message.role == \"assistant\" and message.tool_calls is defined and message.tool_calls is iterable and message.tool_calls | length > 0 %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content is defined and message.content is string and message.content | trim | length > 0 %}\n            {{- '\\n' + message.content | trim + '\\n' }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n<function=' + tool_call.name + '>\\n' }}\n            {%- if tool_call.arguments is defined %}\n                {%- for args_name, args_value in tool_call.arguments|items %}\n                    {{- '<parameter=' + args_name + '>\\n' }}\n                    {%- set args_value = args_value if args_value is string else args_value | string %}\n                    {{- args_value }}\n                    {{- '\\n</parameter>\\n' }}\n                {%- endfor %}\n            {%- endif %}\n            {{- '</function>\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"user\" or message.role == \"system\" or message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.previtem and loop.previtem.role != \"tool\" %}\n            {{- '<|im_start|>user\\n' }}\n        {%- endif %}\n        {{- '<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>\\n' }}\n        {%- if not loop.last and loop.nextitem.role != \"tool\" %}\n            {{- '<|im_end|>\\n' }}\n        {%- elif loop.last %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- else %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+    "clean_up_tokenization_spaces": false,
+    "eos_token": "<|im_end|>",
+    "errors": "replace",
+    "model_max_length": 1048576,
+    "pad_token": "<|endoftext|>",
+    "split_special_tokens": false,
+    "tokenizer_class": "Qwen2Tokenizer",
+    "unk_token": null,
+    "add_bos_token": false
+}