Upload KORMoMoeForCausalLM

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

config.json

@@ -0,0 +1,43 @@
+{
+  "architectures": [
+    "KORMoMoeForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_kormo_moe.KORMoMoeConfig",
+    "AutoModelForCausalLM": "modeling_kormo_moe.KORMoMoeForCausalLM"
+  },
+  "bos_token_id": 125030,
+  "decoder_sparse_step": 1,
+  "dtype": "bfloat16",
+  "eos_token_id": 125040,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 16384,
+  "mask_type": null,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "kormo_moe",
+  "moe_intermediate_size": 16384,
+  "norm_topk_prob": true,
+  "num_attention_heads": 32,
+  "num_experts": 2,
+  "num_experts_per_tok": 2,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 8,
+  "pad_token_id": 125032,
+  "pretrain_tp": 1,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 8000000.0,
+  "shared_expert_intermediate_size": null,
+  "tie_word_embeddings": false,
+  "tie_word_embeddins": false,
+  "transformers_version": "4.57.0",
+  "use_cache": true,
+  "vocab_size": 125184
+}

configuration_kormo_moe.py

@@ -0,0 +1,86 @@
+# <저장된_모델_경로>/configuration_kormo_moe.py
+
+from transformers import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+
+
+class KORMoMoeConfig(PretrainedConfig):
+    model_type = "kormo_moe"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    
+    def __init__(
+        self,
+        vocab_size=112576,
+        hidden_size=6144,
+        intermediate_size=21504,
+        num_hidden_layers=48,
+        num_attention_heads=40,
+        num_key_value_heads=8,
+        hidden_act="silu",
+        max_position_embeddings=131072,
+        initializer_range=0.02,
+        rms_norm_eps=1e-05,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=1,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=500000.0,
+        attention_bias=False,
+        attention_dropout=0.0,
+        rope_scaling=None,
+        mlp_bias=False,
+        head_dim=128,
+        # MoE specific
+        num_experts=2,
+        num_experts_per_tok=2,
+        moe_intermediate_size=None,
+        shared_expert_intermediate_size=None,
+        norm_topk_prob=True,
+        decoder_sparse_step=1,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+        
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.pretraining_tp = pretraining_tp
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        self.mlp_bias = mlp_bias
+        self.head_dim = head_dim if head_dim is not None else self.hidden_size // self.num_attention_heads
+        self.mask_type = None
+        
+        # MoE specific
+        self.num_experts = num_experts
+        self.num_experts_per_tok = num_experts_per_tok
+        self.moe_intermediate_size = moe_intermediate_size if moe_intermediate_size is not None else intermediate_size
+        self.shared_expert_intermediate_size = shared_expert_intermediate_size
+        self.norm_topk_prob = norm_topk_prob
+        self.decoder_sparse_step = decoder_sparse_step
+        
+        if self.rope_scaling is not None and "type" in self.rope_scaling:
+            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
+        rope_config_validation(self)
+        
+        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,
+        )

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 125030,
+  "eos_token_id": 125040,
+  "pad_token_id": 125032,
+  "transformers_version": "4.57.0"
+}

model-00001-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c1dfa9cefb050062dba0d9a2dada46857381a60c7ca3f3acfdf60e6f009d139
+size 4934753320

model-00002-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91f3d80afe8bc8aabb757f98e27055b44edf75729182932cb7cd3a4bedc283ca
+size 4983021920

model-00003-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b16a41b26b9f43f3d75bf093ce58877a9704b028aa03e732fe3628de7b0dea8f
+size 4932690560

model-00004-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f7021347a7d01814ed0dfad2802cbae0ff7f08631c3744654677991952349a9
+size 4932707184

model-00005-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e44ed78f1682fb3ef95b784aeebe7faae8a670c2b6b8c1d04f074a51695dbe4c
+size 4983005368

model-00006-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9acc30f653240e4445f17dd01f2bbd032a204628a7cd3932360f7e92e7a38c5
+size 4932707184

model-00007-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b56a5a079335c2573b0e74e5e14b4f8997c71f7a051171dc7ffaa03a5ca3a7fa
+size 4983005368

model-00008-of-00008.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6aa028f48047eb405328ee1fba2f346c6e2511c060788424fd0484089eb55446
+size 2938203792

model.safetensors.index.json

@@ -0,0 +1,531 @@
+{
+  "metadata": {
+    "total_parameters": 18810015744,
+    "total_size": 37620031488
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00008-of-00008.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.experts.0.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.experts.0.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.experts.0.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.experts.1.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.experts.1.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.experts.1.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.mlp.gate.linear.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.pre_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.pre_mlp_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.experts.0.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.experts.0.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.experts.0.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.experts.1.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.experts.1.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.experts.1.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.mlp.gate.linear.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.pre_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.pre_mlp_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.10.mlp.experts.0.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.mlp.experts.0.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.mlp.experts.0.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.mlp.experts.1.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.mlp.experts.1.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.mlp.experts.1.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.mlp.gate.linear.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.pre_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.pre_mlp_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.experts.0.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.experts.0.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.experts.0.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.experts.1.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.experts.1.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.experts.1.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.mlp.gate.linear.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.pre_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.pre_mlp_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.experts.0.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.experts.0.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.experts.0.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.experts.1.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.experts.1.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.experts.1.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.mlp.gate.linear.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.pre_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.pre_mlp_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.experts.0.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.experts.0.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.experts.0.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.experts.1.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.experts.1.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.experts.1.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.mlp.gate.linear.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.pre_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.pre_mlp_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.experts.0.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.experts.0.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.experts.0.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.experts.1.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.experts.1.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.experts.1.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.mlp.gate.linear.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.pre_attention_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.pre_mlp_layernorm.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.mlp.experts.0.down_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.mlp.experts.0.gate_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.mlp.experts.0.up_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.mlp.experts.1.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.15.mlp.experts.1.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.15.mlp.experts.1.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.15.mlp.gate.linear.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.pre_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.15.pre_mlp_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
+    "model.layers.16.mlp.experts.0.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.mlp.experts.0.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.mlp.experts.0.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.mlp.experts.1.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.mlp.experts.1.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.mlp.experts.1.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.mlp.gate.linear.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.pre_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.pre_mlp_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.experts.0.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.experts.0.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.experts.0.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.experts.1.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.experts.1.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.experts.1.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.mlp.gate.linear.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.pre_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.pre_mlp_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.experts.0.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.experts.0.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.experts.0.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.experts.1.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.experts.1.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.experts.1.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.mlp.gate.linear.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.pre_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.pre_mlp_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.experts.0.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.experts.0.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.experts.0.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.experts.1.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.experts.1.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.experts.1.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.mlp.gate.linear.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.pre_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.pre_mlp_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.2.mlp.experts.0.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.mlp.experts.0.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.mlp.experts.0.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.mlp.experts.1.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.mlp.experts.1.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.mlp.experts.1.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.mlp.gate.linear.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.pre_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.pre_mlp_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.20.mlp.experts.0.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.mlp.experts.0.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.mlp.experts.0.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.mlp.experts.1.down_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.mlp.experts.1.gate_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.mlp.experts.1.up_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.mlp.gate.linear.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.pre_attention_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.pre_mlp_layernorm.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.21.mlp.experts.0.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.mlp.experts.0.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.mlp.experts.0.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.mlp.experts.1.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.mlp.experts.1.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.mlp.experts.1.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.mlp.gate.linear.weight": "model-00004-of-00008.safetensors",
+    "model.layers.21.pre_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.pre_mlp_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
+    "model.layers.22.mlp.experts.0.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.mlp.experts.0.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.mlp.experts.0.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.mlp.experts.1.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.mlp.experts.1.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.mlp.experts.1.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.mlp.gate.linear.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.pre_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.pre_mlp_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.experts.0.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.experts.0.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.experts.0.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.experts.1.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.experts.1.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.experts.1.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.mlp.gate.linear.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.pre_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.pre_mlp_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.experts.0.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.experts.0.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.experts.0.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.experts.1.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.experts.1.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.experts.1.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.mlp.gate.linear.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.pre_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.pre_mlp_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.experts.0.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.experts.0.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.experts.0.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.experts.1.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.experts.1.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.experts.1.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.mlp.gate.linear.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.pre_attention_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.pre_mlp_layernorm.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.mlp.experts.0.down_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.mlp.experts.0.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.mlp.experts.0.up_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.mlp.experts.1.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.26.mlp.experts.1.gate_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.mlp.experts.1.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.26.mlp.gate.linear.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.pre_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.26.pre_mlp_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
+    "model.layers.27.mlp.experts.0.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.mlp.experts.0.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.mlp.experts.0.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.mlp.experts.1.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.mlp.experts.1.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.mlp.experts.1.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.mlp.gate.linear.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.pre_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.pre_mlp_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.experts.0.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.experts.0.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.experts.0.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.experts.1.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.experts.1.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.experts.1.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.mlp.gate.linear.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.pre_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.pre_mlp_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.experts.0.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.experts.0.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.experts.0.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.experts.1.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.experts.1.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.experts.1.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.mlp.gate.linear.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.pre_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.pre_mlp_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.3.mlp.experts.0.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.mlp.experts.0.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.mlp.experts.0.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.mlp.experts.1.down_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.mlp.experts.1.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.mlp.experts.1.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.mlp.gate.linear.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.pre_attention_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.pre_mlp_layernorm.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.30.mlp.experts.0.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.mlp.experts.0.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.mlp.experts.0.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.mlp.experts.1.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.mlp.experts.1.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.mlp.experts.1.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.mlp.gate.linear.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.pre_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.pre_mlp_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.experts.0.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.experts.0.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.experts.0.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.experts.1.down_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.experts.1.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.experts.1.up_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.mlp.gate.linear.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.pre_attention_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.pre_mlp_layernorm.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.32.mlp.experts.0.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.mlp.experts.0.gate_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.32.mlp.experts.0.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.mlp.experts.1.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.mlp.experts.1.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.mlp.experts.1.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.mlp.gate.linear.weight": "model-00006-of-00008.safetensors",
+    "model.layers.32.pre_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.pre_mlp_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
+    "model.layers.33.mlp.experts.0.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.mlp.experts.0.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.mlp.experts.0.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.mlp.experts.1.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.mlp.experts.1.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.mlp.experts.1.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.mlp.gate.linear.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.pre_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.pre_mlp_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.experts.0.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.experts.0.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.experts.0.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.experts.1.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.experts.1.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.experts.1.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.mlp.gate.linear.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.pre_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.pre_mlp_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.experts.0.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.experts.0.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.experts.0.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.experts.1.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.experts.1.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.experts.1.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.mlp.gate.linear.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.pre_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.pre_mlp_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.experts.0.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.experts.0.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.experts.0.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.experts.1.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.experts.1.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.experts.1.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.mlp.gate.linear.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.pre_attention_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.pre_mlp_layernorm.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.mlp.experts.0.down_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.mlp.experts.0.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.mlp.experts.0.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.mlp.experts.1.down_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.37.mlp.experts.1.gate_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.mlp.experts.1.up_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.mlp.gate.linear.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.pre_attention_layernorm.weight": "model-00008-of-00008.safetensors",
+    "model.layers.37.pre_mlp_layernorm.weight": "model-00008-of-00008.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
+    "model.layers.38.mlp.experts.0.down_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.mlp.experts.0.gate_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.mlp.experts.0.up_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.mlp.experts.1.down_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.mlp.experts.1.gate_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.mlp.experts.1.up_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.mlp.gate.linear.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.pre_attention_layernorm.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.pre_mlp_layernorm.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.experts.0.down_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.experts.0.gate_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.experts.0.up_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.experts.1.down_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.experts.1.gate_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.experts.1.up_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.mlp.gate.linear.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.pre_attention_layernorm.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.pre_mlp_layernorm.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00008-of-00008.safetensors",
+    "model.layers.4.mlp.experts.0.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.4.mlp.experts.0.gate_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.4.mlp.experts.0.up_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.4.mlp.experts.1.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.4.mlp.experts.1.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.4.mlp.experts.1.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.4.mlp.gate.linear.weight": "model-00001-of-00008.safetensors",
+    "model.layers.4.pre_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.4.pre_mlp_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00008.safetensors",
+    "model.layers.5.mlp.experts.0.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.mlp.experts.0.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.mlp.experts.0.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.mlp.experts.1.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.mlp.experts.1.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.mlp.experts.1.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.mlp.gate.linear.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.pre_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.pre_mlp_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.experts.0.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.experts.0.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.experts.0.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.experts.1.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.experts.1.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.experts.1.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.mlp.gate.linear.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.pre_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.pre_mlp_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.experts.0.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.experts.0.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.experts.0.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.experts.1.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.experts.1.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.experts.1.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.mlp.gate.linear.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.pre_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.pre_mlp_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.experts.0.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.experts.0.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.experts.0.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.experts.1.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.experts.1.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.experts.1.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.mlp.gate.linear.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.pre_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.pre_mlp_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.experts.0.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.experts.0.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.experts.0.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.experts.1.down_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.experts.1.gate_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.experts.1.up_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.mlp.gate.linear.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.pre_attention_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.pre_mlp_layernorm.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00008.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00008.safetensors",
+    "model.norm.weight": "model-00008-of-00008.safetensors"
+  }
+}

modeling_kormo_moe.py

@@ -0,0 +1,574 @@
+from typing import Callable, List, Optional, Tuple, Union, Dict
+import torch
+from torch import 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.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import can_return_tuple, logging
+from .configuration_kormo_moe import KORMoMoeConfig
+
+logger = logging.get_logger(__name__)
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class RMSNorm(nn.Module):
+    """KORMoRMSNorm is equivalent to T5LayerNorm"""
+    def __init__(self, hidden_size: int, eps: float = 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}"
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs,
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def 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.to(q.dtype), k_embed.to(k.dtype)
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+class Attention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: KORMoMoeConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+        
+        self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_value: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+            
+        return attn_output, attn_weights
+
+
+@use_kernel_forward_from_hub("MLP")
+class MLP(nn.Module):
+    """Basic MLP for experts"""
+    def __init__(self, config, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = intermediate_size if intermediate_size is not None else config.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 MoEGate(nn.Module):
+    """MoE Gating mechanism"""
+    def __init__(self, config: KORMoMoeConfig):
+        super().__init__()
+        self.config = config
+        self.top_k = config.num_experts_per_tok
+        self.n_routed_experts = config.num_experts
+        self.norm_topk_prob = config.norm_topk_prob
+
+        self.linear = nn.Linear(config.hidden_size, config.num_experts, bias=False)
+
+    def forward(self, hidden_states: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+        # hidden_states: [batch_size, seq_len, hidden_size]
+        batch_size, seq_len, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+
+        # Compute router logits
+        router_logits = self.linear(hidden_states)  # [batch_size * seq_len, num_experts]
+        
+        # Get routing weights
+        routing_weights = F.softmax(router_logits, dim=-1, dtype=torch.float)
+        routing_weights, selected_experts = torch.topk(routing_weights, self.top_k, dim=-1)
+        
+        # Normalize routing weights if needed
+        if self.norm_topk_prob:
+            routing_weights = routing_weights / routing_weights.sum(dim=-1, keepdim=True)
+        
+        routing_weights = routing_weights.to(hidden_states.dtype)
+        
+        return routing_weights, selected_experts
+
+
+class KORMoSparseMoeBlock(nn.Module):
+    """KORMo Sparse MoE Block"""
+    def __init__(self, config: KORMoMoeConfig):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.num_experts = config.num_experts
+        self.top_k = config.num_experts_per_tok
+        
+        self.gate = MoEGate(config)
+        self.experts = nn.ModuleList([
+            MLP(config, intermediate_size=config.moe_intermediate_size)
+            for _ in range(self.num_experts)
+        ])
+        
+        # Shared expert (선택사항)
+        self.shared_expert = None
+        self.shared_expert_gate = None
+        if config.shared_expert_intermediate_size is not None:
+            self.shared_expert = MLP(config, intermediate_size=config.shared_expert_intermediate_size)
+            self.shared_expert_gate = nn.Linear(config.hidden_size, 1, bias=False)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        batch_size, seq_len, hidden_dim = hidden_states.shape
+        hidden_states_flat = hidden_states.view(-1, hidden_dim)
+        
+        routing_weights, selected_experts = self.gate(hidden_states)
+        final_hidden_states = torch.zeros_like(hidden_states_flat)
+        
+        expert_mask = torch.nn.functional.one_hot(selected_experts, num_classes=self.num_experts).permute(2, 1, 0)
+        
+        for expert_idx in range(self.num_experts):
+            expert_layer = self.experts[expert_idx]
+            idx, top_x = torch.where(expert_mask[expert_idx])
+            
+            if top_x.shape[0] == 0:
+                continue
+            
+            top_x_list = top_x.tolist()
+            idx_list = idx.tolist()
+            
+            current_state = hidden_states_flat[None, top_x_list].reshape(-1, hidden_dim)
+            current_hidden_states = expert_layer(current_state) * routing_weights[top_x_list, idx_list, None]
+            
+            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
+        
+        final_hidden_states = final_hidden_states.reshape(batch_size, seq_len, hidden_dim)
+        
+        # Shared expert 추가
+        if self.shared_expert is not None:
+            hidden_states_flat = hidden_states.view(-1, hidden_dim)
+            shared_output = self.shared_expert(hidden_states_flat)
+            shared_gate = torch.sigmoid(self.shared_expert_gate(hidden_states_flat))
+            final_hidden_states = final_hidden_states + (shared_gate * shared_output).reshape(batch_size, seq_len, hidden_dim)
+        
+        return final_hidden_states
+
+
+class DecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: KORMoMoeConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = Attention(config=config, layer_idx=layer_idx)
+        self.mlp = KORMoSparseMoeBlock(config)
+        self.pre_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.pre_mlp_layernorm = RMSNorm(config.hidden_size, eps=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_value: Optional[Cache] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        residual = hidden_states
+        hidden_states = self.pre_attention_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # MoE layer
+        residual = hidden_states
+        hidden_states = self.pre_mlp_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        return outputs
+
+
+class RotaryEmbedding(nn.Module):
+    def __init__(self, config: KORMoMoeConfig, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and config.rope_scaling is not None:
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @torch.no_grad()
+    @dynamic_rope_update
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            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, sin
+
+
+class KORMoMoePreTrainedModel(PreTrainedModel):
+    config_class = KORMoMoeConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["DecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn_3 = True
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _supports_cache_class = True
+    _supports_quantized_cache = True
+    _supports_static_cache = True
+    _supports_attention_backend = True
+
+    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, RMSNorm):
+            module.weight.data.fill_(1.0)
+
+
+class KORMoMoeModel(KORMoMoePreTrainedModel):
+    def __init__(self, config: KORMoMoeConfig):
+        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(
+            [DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = RotaryEmbedding(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
+
+    @can_return_tuple
+    def forward(
+        self,
+        input_ids: 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,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **flash_attn_kwargs: Unpack[FlashAttentionKwargs],
+    ) -> BaseModelOutputWithPast:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        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
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
+            )
+            use_cache = False
+
+        if not isinstance(past_key_values, (type(None), Cache)):
+            raise ValueError("The `past_key_values` should be either a `Cache` object or `None`.")
+        
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **flash_attn_kwargs,
+            )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values if use_cache else None,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+    
+class KORMoMoeForCausalLM(KORMoMoePreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = KORMoMoeModel(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
+
+    @can_return_tuple
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: int = 0,
+        **kwargs,
+    ) -> CausalLMOutputWithPast:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+
+        outputs: BaseModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )