Upload folder using huggingface_hub

.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

chat_template.jinja

@@ -0,0 +1,86 @@
+[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) }}
+{%- 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 ((clear_thinking is defined and not clear_thinking) or loop.index0 > ns.last_user_index) and reasoning_content -%}
+{{ '<think>' + reasoning_content.strip() +  '</think>'}}
+{%- else -%}
+{{ '</think>' }}
+{%- endif -%}
+{%- if content.strip() -%}
+{{ content.strip() }}
+{%- endif -%}
+{% if m.tool_calls %}
+{% for tc in m.tool_calls %}
+{%- if tc.function %}
+    {%- set tc = tc.function %}
+{%- endif %}
+{{- '<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 %}
+{{- '<tool_response>' }}
+{{- m.content }}
+{{- '</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|>{{- '</think>' if (enable_thinking is defined and not enable_thinking) else '<think>' -}}
+{%- endif -%}

config.json

@@ -0,0 +1,108 @@
+{
+  "architectures": [
+    "Glm4MoeLiteSCMForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_glm_scm.Glm4MoeLiteSCMConfig",
+    "AutoModel": "modeling_glm_scm.Glm4MoeLiteSCMModel",
+    "AutoModelForCausalLM": "modeling_glm_scm.Glm4MoeLiteSCMForCausalLM"
+  },
+  "bos_token_id": 0,
+  "dtype": "bfloat16",
+  "eos_token_id": [
+    154820,
+    154827,
+    154829
+  ],
+  "first_k_dense_replace": 1,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 10240,
+  "kv_lora_rank": 512,
+  "max_position_embeddings": 202752,
+  "mlp_layer_types": [
+    "dense",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse",
+    "sparse"
+  ],
+  "model_type": "glm4_moe_lite",
+  "moe_intermediate_size": 1536,
+  "n_group": 1,
+  "n_routed_experts": 64,
+  "n_shared_experts": 1,
+  "norm_topk_prob": true,
+  "num_attention_heads": 20,
+  "num_experts_per_tok": 4,
+  "num_hidden_layers": 47,
+  "num_key_value_heads": 20,
+  "num_nextn_predict_layers": 1,
+  "pad_token_id": 154820,
+  "q_lora_rank": 768,
+  "qk_head_dim": 256,
+  "qk_nope_head_dim": 192,
+  "qk_rope_head_dim": 64,
+  "rms_norm_eps": 1e-05,
+  "rope_interleave": true,
+  "rope_parameters": {
+    "rope_theta": 10000.0,
+    "rope_type": "default"
+  },
+  "rope_theta": 1000000,
+  "routed_scaling_factor": 1.8,
+  "scoring_func": "sigmoid",
+  "tie_word_embeddings": false,
+  "topk_group": 1,
+  "topk_method": "noaux_tc",
+  "transformers_version": "5.0.0",
+  "use_cache": true,
+  "v_head_dim": 256,
+  "vocab_size": 154880
+}

configuration_glm_scm.py

@@ -0,0 +1,110 @@
+# Configuration for GLM-4.7-Flash ScatterMoE (SCM) variant
+# Based on Glm4MoeLiteConfig from transformers
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class Glm4MoeLiteSCMConfig(PretrainedConfig):
+    model_type = "glm4_moe_lite"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=154880,
+        hidden_size=2048,
+        intermediate_size=10240,
+        moe_intermediate_size=1536,
+        num_hidden_layers=47,
+        num_attention_heads=20,
+        num_key_value_heads=20,
+        n_shared_experts=1,
+        n_routed_experts=64,
+        routed_scaling_factor=1.8,
+        kv_lora_rank=512,
+        q_lora_rank=768,
+        qk_rope_head_dim=64,
+        v_head_dim=256,
+        qk_nope_head_dim=192,
+        n_group=1,
+        topk_group=1,
+        num_experts_per_tok=4,
+        norm_topk_prob=True,
+        topk_method="noaux_tc",
+        first_k_dense_replace=1,
+        num_nextn_predict_layers=1,
+        hidden_act="silu",
+        max_position_embeddings=202752,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=1,
+        tie_word_embeddings=False,
+        rope_theta=1000000,
+        rope_scaling=None,
+        rope_interleave=True,
+        attention_bias=False,
+        attention_dropout=0.0,
+        scoring_func="sigmoid",
+        mlp_layer_types=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.moe_intermediate_size = moe_intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.n_shared_experts = n_shared_experts
+        self.n_routed_experts = n_routed_experts
+        self.routed_scaling_factor = routed_scaling_factor
+        self.kv_lora_rank = kv_lora_rank
+        self.q_lora_rank = q_lora_rank
+        self.qk_rope_head_dim = qk_rope_head_dim
+        self.v_head_dim = v_head_dim
+        self.qk_nope_head_dim = qk_nope_head_dim
+        self.qk_head_dim = qk_nope_head_dim + qk_rope_head_dim
+        self.head_dim = qk_rope_head_dim  # Used for RoPE computation
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.num_experts_per_tok = num_experts_per_tok
+        self.norm_topk_prob = norm_topk_prob
+        self.topk_method = topk_method
+        self.first_k_dense_replace = first_k_dense_replace
+        self.num_nextn_predict_layers = num_nextn_predict_layers
+        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.rope_interleave = rope_interleave
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        self.scoring_func = scoring_func
+
+        # MLP layer types: first layer is dense, rest are sparse (MoE)
+        if mlp_layer_types is not None:
+            self.mlp_layer_types = mlp_layer_types
+        else:
+            self.mlp_layer_types = (
+                ["dense"] * first_k_dense_replace
+                + ["sparse"] * (num_hidden_layers - first_k_dense_replace)
+            )
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+
+__all__ = ["Glm4MoeLiteSCMConfig"]

generation_config.json

@@ -0,0 +1,14 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": [
+    154820,
+    154827,
+    154829
+  ],
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 154820,
+  "transformers_version": "5.0.0",
+  "use_cache": true
+}

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68f662181d6d33cb19b05d929b056d77b569cb68e802fdaeb0b1041b484a8e6b
+size 49722148632

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c00c9624be6b5b7fb17864350302f91bf54db6ae93e0a85add59583c67ab500
+size 10164736704

model.safetensors.index.json

@@ -0,0 +1,759 @@
+{
+  "metadata": {
+    "total_parameters": 29943390976,
+    "total_size": 59886793728
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00001-of-00002.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.0.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.1.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.10.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.11.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.12.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.13.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.14.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.15.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.16.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.17.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.18.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.19.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.2.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.20.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.21.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.22.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.23.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.24.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.25.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.26.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.27.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.28.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.29.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.3.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.30.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.31.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.32.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.33.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.34.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.35.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.36.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.37.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.38.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.39.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.39.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.39.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.39.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.4.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.40.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.40.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.41.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.42.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.43.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.44.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.45.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.input_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.gate.e_score_correction_bias": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.gate.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.moe_mlp.experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.moe_mlp.output_experts.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.shared_experts.down_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.shared_experts.gate_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.mlp.shared_experts.up_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.kv_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.kv_a_proj_with_mqa.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.kv_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.q_a_layernorm.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.q_a_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.46.self_attn.q_b_proj.weight": "model-00002-of-00002.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.5.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.6.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.7.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.8.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate.e_score_correction_bias": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.gate.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.moe_mlp.experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.moe_mlp.output_experts.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.shared_experts.down_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.shared_experts.gate_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.mlp.shared_experts.up_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.kv_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.kv_a_proj_with_mqa.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.kv_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_a_layernorm.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_a_proj.weight": "model-00001-of-00002.safetensors",
+    "model.layers.9.self_attn.q_b_proj.weight": "model-00001-of-00002.safetensors",
+    "model.norm.weight": "model-00002-of-00002.safetensors"
+  }
+}

modeling_glm_scm.py

@@ -0,0 +1,651 @@
+# coding=utf-8
+# ScatterMoE (SCM) variant of GLM-4.7-Flash (Glm4MoeLite architecture)
+# Based on transformers/models/glm4_moe_lite/modeling_glm4_moe_lite.py
+# with fused expert kernels via scattermoe
+#
+# Copyright 2025 the HuggingFace Team. All rights reserved.
+# Licensed under the Apache License, Version 2.0
+
+import math
+from typing import Optional, Tuple, Union, List
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+
+import scattermoe
+
+try:
+    from .configuration_glm_scm import Glm4MoeLiteSCMConfig
+except ImportError:
+    from configuration_glm_scm import Glm4MoeLiteSCMConfig
+
+
+logger = logging.get_logger(__name__)
+
+
+class Glm4MoeLiteSCMRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        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 Glm4MoeLiteSCMRotaryEmbedding(nn.Module):
+    def __init__(self, config: Glm4MoeLiteSCMConfig, device=None):
+        super().__init__()
+        self.config = config
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        # Compute inverse frequencies
+        base = config.rope_theta
+        head_dim = config.head_dim  # qk_rope_head_dim
+        dim = head_dim  # partial_rotary_factor = 1.0
+        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2, dtype=torch.int64).to(device=device, dtype=torch.float) / dim))
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)
+        self.attention_scaling = 1.0
+
+    @torch.no_grad()
+    def forward(self, x, position_ids=None):
+        if position_ids is None:
+            # Fallback
+            seq_len = x.shape[1] if x.dim() > 2 else x.shape[0]
+            position_ids = torch.arange(seq_len, device=x.device).unsqueeze(0)
+
+        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):
+            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)
+
+
+def rotate_half(x):
+    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, unsqueeze_dim=1):
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def apply_rotary_pos_emb_interleave(q, k, cos, sin, unsqueeze_dim=1):
+    """Interleaved RoPE: reorder dimensions before applying standard rotation."""
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+
+    b, h, s, d = q.shape
+    q = q.view(b, h, s, d // 2, 2).transpose(4, 3).reshape(b, h, s, d)
+    b, h, s, d = k.shape
+    k = k.view(b, h, s, d // 2, 2).transpose(4, 3).reshape(b, h, s, d)
+
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    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 yarn_get_mscale(scale=1, mscale=1):
+    if scale <= 1:
+        return 1.0
+    return 0.1 * mscale * math.log(scale) + 1.0
+
+
+class Glm4MoeLiteSCMAttention(nn.Module):
+    """Multi-head Latent Attention (MLA) with LoRA-compressed KV cache."""
+
+    def __init__(self, config: Glm4MoeLiteSCMConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.attention_dropout = config.attention_dropout
+        self.num_heads = config.num_attention_heads
+
+        self.q_lora_rank = config.q_lora_rank
+        self.qk_rope_head_dim = config.qk_rope_head_dim
+        self.kv_lora_rank = config.kv_lora_rank
+        self.v_head_dim = config.v_head_dim
+        self.qk_nope_head_dim = config.qk_nope_head_dim
+        self.qk_head_dim = config.qk_head_dim
+
+        self.is_causal = True
+
+        if self.q_lora_rank is None:
+            self.q_proj = nn.Linear(config.hidden_size, self.num_heads * self.qk_head_dim, bias=False)
+        else:
+            self.q_a_proj = nn.Linear(config.hidden_size, config.q_lora_rank, bias=config.attention_bias)
+            self.q_a_layernorm = Glm4MoeLiteSCMRMSNorm(config.q_lora_rank)
+            self.q_b_proj = nn.Linear(config.q_lora_rank, self.num_heads * self.qk_head_dim, bias=False)
+
+        self.kv_a_proj_with_mqa = nn.Linear(
+            config.hidden_size, self.kv_lora_rank + self.qk_rope_head_dim, bias=config.attention_bias,
+        )
+        self.kv_a_layernorm = Glm4MoeLiteSCMRMSNorm(self.kv_lora_rank)
+        self.kv_b_proj = nn.Linear(
+            self.kv_lora_rank, self.num_heads * (self.qk_nope_head_dim + self.v_head_dim), bias=False,
+        )
+        self.o_proj = nn.Linear(self.num_heads * self.v_head_dim, config.hidden_size, bias=config.attention_bias)
+
+        self.scaling = self.qk_head_dim ** (-0.5)
+        if config.rope_scaling is not None and isinstance(config.rope_scaling, dict) and "factor" in config.rope_scaling:
+            mscale_all_dim = config.rope_scaling.get("mscale_all_dim", 0)
+            scaling_factor = config.rope_scaling["factor"]
+            if mscale_all_dim:
+                mscale = yarn_get_mscale(scaling_factor, mscale_all_dim)
+                self.scaling = self.scaling * mscale * mscale
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: Tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
+        batch_size, seq_length = hidden_states.shape[:-1]
+        query_shape = (batch_size, seq_length, -1, self.qk_head_dim)
+        key_shape = (batch_size, seq_length, -1, self.qk_nope_head_dim + self.v_head_dim)
+
+        if self.q_lora_rank is None:
+            q_states = self.q_proj(hidden_states)
+        else:
+            q_states = self.q_b_proj(self.q_a_layernorm(self.q_a_proj(hidden_states)))
+        q_states = q_states.view(query_shape).transpose(1, 2)
+        q_pass, q_rot = torch.split(q_states, [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
+
+        compressed_kv = self.kv_a_proj_with_mqa(hidden_states)
+        k_pass, k_rot = torch.split(compressed_kv, [self.kv_lora_rank, self.qk_rope_head_dim], dim=-1)
+
+        k_pass = self.kv_b_proj(self.kv_a_layernorm(k_pass)).view(key_shape).transpose(1, 2)
+        k_pass, value_states = torch.split(k_pass, [self.qk_nope_head_dim, self.v_head_dim], dim=-1)
+
+        k_rot = k_rot.view(batch_size, 1, seq_length, self.qk_rope_head_dim)
+
+        cos, sin = position_embeddings
+        if getattr(self.config, "rope_interleave", True):
+            q_rot, k_rot = apply_rotary_pos_emb_interleave(q_rot, k_rot, cos, sin)
+        else:
+            q_rot, k_rot = apply_rotary_pos_emb(q_rot, k_rot, cos, sin)
+        k_rot = k_rot.expand(*k_pass.shape[:-1], -1)
+
+        query_states = torch.cat((q_pass, q_rot), dim=-1)
+        key_states = torch.cat((k_pass, k_rot), dim=-1)
+
+        if past_key_values is not None:
+            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)
+
+        # Eager attention
+        key_states_rep = repeat_kv(key_states, self.num_key_value_groups)
+        value_states_rep = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states_rep.transpose(2, 3)) * self.scaling
+        if attention_mask is not None:
+            causal_mask = attention_mask[:, :, :, :key_states_rep.shape[-2]]
+            attn_weights = attn_weights + causal_mask
+
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_weights = nn.functional.dropout(attn_weights, p=0.0 if not self.training else self.attention_dropout, training=self.training)
+        attn_output = torch.matmul(attn_weights, value_states_rep)
+        attn_output = attn_output.transpose(1, 2).contiguous()
+
+        attn_output = attn_output.reshape(batch_size, seq_length, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Glm4MoeLiteSCMMLP(nn.Module):
+    def __init__(self, config, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.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):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+class Glm4MoeLiteSCMTopkRouter(nn.Module):
+    def __init__(self, config: Glm4MoeLiteSCMConfig):
+        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))
+        return router_logits
+
+
+class Glm4MoeLiteSCMMoE(nn.Module):
+    """ScatterMoE variant: fused expert computation via Triton kernels."""
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.gate = Glm4MoeLiteSCMTopkRouter(config)
+
+        # ScatterMoE fused experts
+        self.moe_mlp = scattermoe.mlp.GLUMLP(
+            input_size=config.hidden_size,
+            hidden_size=config.moe_intermediate_size,
+            num_experts=config.n_routed_experts,
+            top_k=config.num_experts_per_tok,
+            activation=ACT2FN[config.hidden_act],
+        )
+
+        self.shared_experts = Glm4MoeLiteSCMMLP(
+            config=config, intermediate_size=config.moe_intermediate_size * config.n_shared_experts,
+        )
+
+        self.n_routed_experts = config.n_routed_experts
+        self.n_group = config.n_group
+        self.topk_group = config.topk_group
+        self.norm_topk_prob = config.norm_topk_prob
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.top_k = config.num_experts_per_tok
+
+    def route_tokens_to_experts(self, router_logits):
+        router_logits = router_logits.sigmoid()
+        router_logits_for_choice = router_logits + self.gate.e_score_correction_bias
+        group_scores = (
+            router_logits_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 = router_logits_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]
+        topk_weights = router_logits.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
+
+    def forward(self, hidden_states):
+        residuals = hidden_states
+        orig_shape = hidden_states.shape
+        router_logits = self.gate(hidden_states)
+        topk_indices, topk_weights = self.route_tokens_to_experts(router_logits)
+        hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+
+        # ScatterMoE forward: fused expert computation via Triton kernels
+        hidden_states = self.moe_mlp(hidden_states, topk_weights.to(hidden_states.dtype), topk_indices)
+        hidden_states = hidden_states.view(*orig_shape)
+
+        hidden_states = hidden_states + self.shared_experts(residuals)
+        return hidden_states
+
+
+class Glm4MoeLiteSCMDecoderLayer(nn.Module):
+    def __init__(self, config: Glm4MoeLiteSCMConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = Glm4MoeLiteSCMAttention(config, layer_idx)
+
+        if config.mlp_layer_types[layer_idx] == "sparse":
+            self.mlp = Glm4MoeLiteSCMMoE(config)
+        else:
+            self.mlp = Glm4MoeLiteSCMMLP(config)
+
+        self.input_layernorm = Glm4MoeLiteSCMRMSNorm(config.hidden_size, config.rms_norm_eps)
+        self.post_attention_layernorm = Glm4MoeLiteSCMRMSNorm(config.hidden_size, config.rms_norm_eps)
+
+    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,
+        **kwargs,
+    ) -> torch.Tensor:
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+        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
+
+        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
+
+
+class Glm4MoeLiteSCMPreTrainedModel(PreTrainedModel):
+    config_class = Glm4MoeLiteSCMConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Glm4MoeLiteSCMDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+    _keep_in_fp32_modules = ["e_score_correction_bias"]
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, Glm4MoeLiteSCMTopkRouter):
+            module.weight.data.normal_(mean=0.0, std=std)
+            module.e_score_correction_bias.zero_()
+
+
+class Glm4MoeLiteSCMModel(Glm4MoeLiteSCMPreTrainedModel):
+    _keys_to_ignore_on_load_unexpected = [r"model\.layers\.47.*"]
+
+    def __init__(self, config: Glm4MoeLiteSCMConfig):
+        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(
+            [Glm4MoeLiteSCMDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = Glm4MoeLiteSCMRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = Glm4MoeLiteSCMRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    def _prepare_4d_causal_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
+        """Create 4D causal mask."""
+        batch_size, seq_length = input_shape
+        dtype = inputs_embeds.dtype
+        device = inputs_embeds.device
+
+        # Create causal mask
+        if seq_length > 1:
+            causal_mask = torch.full((seq_length, seq_length), torch.finfo(dtype).min, device=device, dtype=dtype)
+            causal_mask = torch.triu(causal_mask, diagonal=1)
+            if past_key_values_length > 0:
+                causal_mask = torch.cat([
+                    torch.zeros(seq_length, past_key_values_length, device=device, dtype=dtype),
+                    causal_mask,
+                ], dim=-1)
+        else:
+            total_length = past_key_values_length + seq_length
+            causal_mask = torch.zeros(1, total_length, device=device, dtype=dtype)
+
+        causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
+
+        if attention_mask is not None:
+            # Expand attention mask
+            expanded_mask = attention_mask[:, None, None, :].to(dtype)
+            expanded_mask = (1.0 - expanded_mask) * torch.finfo(dtype).min
+            total_length = past_key_values_length + seq_length
+            if expanded_mask.shape[-1] < total_length:
+                expanded_mask = F.pad(expanded_mask, (total_length - expanded_mask.shape[-1], 0))
+            causal_mask = causal_mask + expanded_mask
+
+        return causal_mask
+
+    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,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        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 = self.embed_tokens(input_ids)
+
+        batch_size, seq_length = inputs_embeds.shape[:2]
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        past_key_values_length = 0
+        if past_key_values is not None:
+            past_key_values_length = past_key_values.get_seq_length()
+
+        if cache_position is None:
+            cache_position = torch.arange(
+                past_key_values_length, past_key_values_length + seq_length, device=inputs_embeds.device,
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        # Create causal mask
+        causal_mask = self._prepare_4d_causal_attention_mask(
+            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length,
+        )
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids=position_ids)
+
+        all_hidden_states = () if output_hidden_states else None
+
+        for decoder_layer in self.layers[:self.config.num_hidden_layers]:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                hidden_states = torch.utils.checkpoint.checkpoint(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    past_key_values,
+                    use_cache,
+                    cache_position,
+                    position_embeddings,
+                    use_reentrant=False,
+                )
+            else:
+                hidden_states = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_embeddings=position_embeddings,
+                    position_ids=position_ids,
+                    past_key_values=past_key_values,
+                    use_cache=use_cache,
+                    cache_position=cache_position,
+                    **kwargs,
+                )
+
+        hidden_states = self.norm(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, past_key_values, all_hidden_states] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+            hidden_states=all_hidden_states,
+        )
+
+
+class Glm4MoeLiteSCMForCausalLM(Glm4MoeLiteSCMPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = {"lm_head.weight": "model.embed_tokens.weight"}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = Glm4MoeLiteSCMModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    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,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = 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,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            **kwargs,
+        )
+
+        hidden_states = outputs[0]
+        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:
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            loss_fct = nn.CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+        )
+
+
+__all__ = [
+    "Glm4MoeLiteSCMForCausalLM",
+    "Glm4MoeLiteSCMModel",
+    "Glm4MoeLiteSCMPreTrainedModel",
+]

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19e773648cb4e65de8660ea6365e10acca112d42a854923df93db4a6f333a82d
+size 20217442

tokenizer_config.json

@@ -0,0 +1,321 @@
+{
+  "added_tokens_decoder": {
+    "154820": {
+      "content": "<|endoftext|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154821": {
+      "content": "[MASK]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154822": {
+      "content": "[gMASK]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154823": {
+      "content": "[sMASK]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154824": {
+      "content": "<sop>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154825": {
+      "content": "<eop>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154826": {
+      "content": "<|system|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154827": {
+      "content": "<|user|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154828": {
+      "content": "<|assistant|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154829": {
+      "content": "<|observation|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154830": {
+      "content": "<|begin_of_image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154831": {
+      "content": "<|end_of_image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154832": {
+      "content": "<|begin_of_video|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154833": {
+      "content": "<|end_of_video|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154834": {
+      "content": "<|begin_of_audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154835": {
+      "content": "<|end_of_audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154836": {
+      "content": "<|begin_of_transcription|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154837": {
+      "content": "<|end_of_transcription|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "154838": {
+      "content": "<|code_prefix|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154839": {
+      "content": "<|code_middle|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154840": {
+      "content": "<|code_suffix|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154841": {
+      "content": "<think>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154842": {
+      "content": "</think>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154843": {
+      "content": "<tool_call>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154844": {
+      "content": "</tool_call>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154845": {
+      "content": "<tool_response>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154846": {
+      "content": "</tool_response>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154847": {
+      "content": "<arg_key>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154848": {
+      "content": "</arg_key>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154849": {
+      "content": "<arg_value>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154850": {
+      "content": "</arg_value>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154851": {
+      "content": "/nothink",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154852": {
+      "content": "<|begin_of_box|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154853": {
+      "content": "<|end_of_box|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154854": {
+      "content": "<|image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": false
+    },
+    "154855": {
+      "content": "<|video|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": 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|>"
+  ],
+  "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"
+}