diff --git a/docs/source/models/supported_models.rst b/docs/source/models/supported_models.rst
index 745b4b8e2e0eb..9e0303e1dab6c 100644
--- a/docs/source/models/supported_models.rst
+++ b/docs/source/models/supported_models.rst
@@ -127,6 +127,10 @@ Decoder-only Language Models
     - Nemotron-3, Nemotron-4, Minitron
     - :code:`nvidia/Minitron-8B-Base`, :code:`mgoin/Nemotron-4-340B-Base-hf-FP8`, etc.
     - ✅︎
+  * - :code:`OLMoEForCausalLM`
+    - OLMoE
+    - :code:`allenai/OLMoE-1B-7B-0924`, :code:`allenai/OLMoE-1B-7B-0924-Instruct`, etc.
+    -
   * - :code:`OLMoForCausalLM`
     - OLMo
     - :code:`allenai/OLMo-1B-hf`, :code:`allenai/OLMo-7B-hf`, etc.
diff --git a/vllm/model_executor/models/__init__.py b/vllm/model_executor/models/__init__.py
index 7427060922281..bee312a14f440 100644
--- a/vllm/model_executor/models/__init__.py
+++ b/vllm/model_executor/models/__init__.py
@@ -46,6 +46,7 @@
     "MiniCPM3ForCausalLM": ("minicpm3", "MiniCPM3ForCausalLM"),
     "NemotronForCausalLM": ("nemotron", "NemotronForCausalLM"),
     "OlmoForCausalLM": ("olmo", "OlmoForCausalLM"),
+    "OlmoeForCausalLM": ("olmoe", "OlmoeForCausalLM"),
     "OPTForCausalLM": ("opt", "OPTForCausalLM"),
     "OrionForCausalLM": ("orion", "OrionForCausalLM"),
     "PersimmonForCausalLM": ("persimmon", "PersimmonForCausalLM"),
diff --git a/vllm/model_executor/models/olmoe.py b/vllm/model_executor/models/olmoe.py
new file mode 100644
index 0000000000000..c76e5e86c89d8
--- /dev/null
+++ b/vllm/model_executor/models/olmoe.py
@@ -0,0 +1,409 @@
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Inference-only OLMoE model compatible with HuggingFace weights."""
+from typing import Any, Dict, Iterable, List, Optional, Tuple
+
+import torch
+from torch import nn
+from transformers import PretrainedConfig
+
+from vllm.attention import Attention, AttentionMetadata
+from vllm.config import CacheConfig
+from vllm.distributed import get_tensor_model_parallel_world_size
+from vllm.model_executor.layers.fused_moe import FusedMoE
+from vllm.model_executor.layers.layernorm import RMSNorm
+from vllm.model_executor.layers.linear import (QKVParallelLinear,
+                                               ReplicatedLinear,
+                                               RowParallelLinear)
+from vllm.model_executor.layers.logits_processor import LogitsProcessor
+from vllm.model_executor.layers.quantization.base_config import (
+    QuantizationConfig)
+from vllm.model_executor.layers.rotary_embedding import get_rope
+from vllm.model_executor.layers.sampler import Sampler, SamplerOutput
+from vllm.model_executor.layers.vocab_parallel_embedding import (
+    ParallelLMHead, VocabParallelEmbedding)
+from vllm.model_executor.model_loader.weight_utils import default_weight_loader
+from vllm.model_executor.sampling_metadata import SamplingMetadata
+from vllm.sequence import IntermediateTensors
+from vllm.utils import print_warning_once
+
+
+class OlmoeMoE(nn.Module):
+    """A tensor-parallel MoE implementation for Olmoe that shards each expert
+    across all ranks.
+
+    Each expert's weights are sharded across all ranks and a fused MoE
+    kernel is used for the forward pass, and finally we reduce the outputs
+    across ranks.
+    """
+
+    def __init__(self,
+                 num_experts: int,
+                 top_k: int,
+                 hidden_size: int,
+                 intermediate_size: int,
+                 params_dtype: Optional[torch.dtype] = None,
+                 quant_config: Optional[QuantizationConfig] = None,
+                 tp_size: Optional[int] = None,
+                 prefix: str = ""):
+        super().__init__()
+        self.hidden_size = hidden_size
+
+        # Gate always runs at half / full precision for now.
+        self.gate = ReplicatedLinear(hidden_size,
+                                     num_experts,
+                                     bias=False,
+                                     quant_config=None)
+
+        self.experts = FusedMoE(num_experts=num_experts,
+                                top_k=top_k,
+                                hidden_size=hidden_size,
+                                intermediate_size=intermediate_size,
+                                reduce_results=True,
+                                renormalize=False,
+                                quant_config=quant_config,
+                                tp_size=tp_size)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        # NOTE: hidden_states can have either 1D or 2D shape.
+        orig_shape = hidden_states.shape
+        hidden_dim = hidden_states.shape[-1]
+        hidden_states = hidden_states.view(-1, hidden_dim)
+        # router_logits: (num_tokens, n_experts)
+        router_logits, _ = self.gate(hidden_states)
+        final_hidden_states = self.experts(hidden_states=hidden_states,
+                                           router_logits=router_logits)
+        return final_hidden_states.view(orig_shape)
+
+
+class OlmoeAttention(nn.Module):
+
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        num_kv_heads: int,
+        rope_theta: float = 10000,
+        rope_scaling: Optional[Dict[str, Any]] = None,
+        max_position_embeddings: int = 4096,
+        cache_config: Optional[CacheConfig] = None,
+        quant_config: Optional[QuantizationConfig] = None,
+    ) -> None:
+        super().__init__()
+        self.hidden_size = hidden_size
+        tp_size = get_tensor_model_parallel_world_size()
+        self.total_num_heads = num_heads
+        assert self.total_num_heads % tp_size == 0
+        self.num_heads = self.total_num_heads // tp_size
+        self.total_num_kv_heads = num_kv_heads
+        if self.total_num_kv_heads >= tp_size:
+            # Number of KV heads is greater than TP size, so we partition
+            # the KV heads across multiple tensor parallel GPUs.
+            assert self.total_num_kv_heads % tp_size == 0
+        else:
+            # Number of KV heads is less than TP size, so we replicate
+            # the KV heads across multiple tensor parallel GPUs.
+            assert tp_size % self.total_num_kv_heads == 0
+        self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)
+        self.head_dim = hidden_size // self.total_num_heads
+        self.q_size = self.num_heads * self.head_dim
+        self.kv_size = self.num_kv_heads * self.head_dim
+        self.scaling = self.head_dim**-0.5
+        self.rope_theta = rope_theta
+        self.max_position_embeddings = max_position_embeddings
+
+        self.qkv_proj = QKVParallelLinear(
+            hidden_size,
+            self.head_dim,
+            self.total_num_heads,
+            self.total_num_kv_heads,
+            bias=False,
+            quant_config=quant_config,
+        )
+        self.q_norm = RMSNorm(hidden_size, eps=1e-5)
+        self.k_norm = RMSNorm(hidden_size, eps=1e-5)
+        self.o_proj = RowParallelLinear(
+            self.total_num_heads * self.head_dim,
+            hidden_size,
+            bias=False,
+            quant_config=quant_config,
+        )
+
+        self.rotary_emb = get_rope(
+            self.head_dim,
+            rotary_dim=self.head_dim,
+            max_position=max_position_embeddings,
+            base=rope_theta,
+            rope_scaling=rope_scaling,
+            is_neox_style=True,
+        )
+        self.attn = Attention(self.num_heads,
+                              self.head_dim,
+                              self.scaling,
+                              num_kv_heads=self.num_kv_heads,
+                              cache_config=cache_config,
+                              quant_config=quant_config)
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        kv_cache: torch.Tensor,
+        attn_metadata: AttentionMetadata,
+    ) -> torch.Tensor:
+        qkv, _ = self.qkv_proj(hidden_states)
+        q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
+        q, k = self.q_norm(q.contiguous()), self.k_norm(k.contiguous())
+        q, k = self.rotary_emb(positions, q, k)
+        attn_output = self.attn(q, k, v, kv_cache, attn_metadata)
+        output, _ = self.o_proj(attn_output)
+        return output
+
+
+class OlmoeDecoderLayer(nn.Module):
+
+    def __init__(
+        self,
+        config: PretrainedConfig,
+        layer_idx: int,
+        cache_config: Optional[CacheConfig] = None,
+        quant_config: Optional[QuantizationConfig] = None,
+    ) -> None:
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        rope_theta = getattr(config, "rope_theta", 10000)
+        rope_scaling = getattr(config, "rope_scaling", None)
+        max_position_embeddings = getattr(config, "max_position_embeddings",
+                                          4096)
+
+        self.self_attn = OlmoeAttention(
+            hidden_size=self.hidden_size,
+            num_heads=config.num_attention_heads,
+            num_kv_heads=config.num_key_value_heads,
+            rope_theta=rope_theta,
+            rope_scaling=rope_scaling,
+            max_position_embeddings=max_position_embeddings,
+            cache_config=cache_config,
+            quant_config=quant_config,
+        )
+
+        self.mlp = OlmoeMoE(
+            num_experts=config.num_experts,
+            top_k=config.num_experts_per_tok,
+            hidden_size=config.hidden_size,
+            intermediate_size=config.intermediate_size,
+            quant_config=quant_config,
+        )
+        self.input_layernorm = RMSNorm(config.hidden_size, eps=1e-5)
+        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=1e-5)
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        kv_cache: torch.Tensor,
+        attn_metadata: AttentionMetadata,
+        residual: Optional[torch.Tensor],
+    ) -> torch.Tensor:
+        # Self Attention
+        if residual is None:
+            residual = hidden_states
+            hidden_states = self.input_layernorm(hidden_states)
+        else:
+            hidden_states, residual = self.input_layernorm(
+                hidden_states, residual)
+
+        hidden_states = self.self_attn(
+            positions=positions,
+            hidden_states=hidden_states,
+            kv_cache=kv_cache,
+            attn_metadata=attn_metadata,
+        )
+
+        # Fully Connected
+        hidden_states, residual = self.post_attention_layernorm(
+            hidden_states, residual)
+        hidden_states = self.mlp(hidden_states)
+        return hidden_states, residual
+
+
+class OlmoeModel(nn.Module):
+
+    def __init__(
+        self,
+        config: PretrainedConfig,
+        cache_config: Optional[CacheConfig] = None,
+        quant_config: Optional[QuantizationConfig] = None,
+    ) -> None:
+        super().__init__()
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = VocabParallelEmbedding(
+            config.vocab_size,
+            config.hidden_size,
+        )
+        self.layers = nn.ModuleList([
+            OlmoeDecoderLayer(config,
+                              layer_idx,
+                              cache_config,
+                              quant_config=quant_config)
+            for layer_idx in range(config.num_hidden_layers)
+        ])
+        self.norm = RMSNorm(config.hidden_size, eps=1e-5)
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        kv_caches: List[torch.Tensor],
+        attn_metadata: AttentionMetadata,
+    ) -> torch.Tensor:
+        hidden_states = self.embed_tokens(input_ids)
+        residual = None
+        for i in range(len(self.layers)):
+            layer = self.layers[i]
+            hidden_states, residual = layer(positions, hidden_states,
+                                            kv_caches[i], attn_metadata,
+                                            residual)
+        hidden_states, _ = self.norm(hidden_states, residual)
+        return hidden_states
+
+
+class OlmoeForCausalLM(nn.Module):
+
+    fall_back_to_pt_during_load = False
+
+    def __init__(
+        self,
+        config: PretrainedConfig,
+        cache_config: Optional[CacheConfig] = None,
+        quant_config: Optional[QuantizationConfig] = None,
+    ) -> None:
+        super().__init__()
+        self.config = config
+        self.quant_config = quant_config
+        self.model = OlmoeModel(config, cache_config, quant_config)
+        self.lm_head = ParallelLMHead(config.vocab_size,
+                                      config.hidden_size,
+                                      quant_config=quant_config)
+        self.logits_processor = LogitsProcessor(config.vocab_size)
+        self.sampler = Sampler()
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        kv_caches: List[torch.Tensor],
+        attn_metadata: AttentionMetadata,
+        intermediate_tensors: Optional[IntermediateTensors] = None,
+    ) -> torch.Tensor:
+        hidden_states = self.model(input_ids, positions, kv_caches,
+                                   attn_metadata)
+        return hidden_states
+
+    def compute_logits(self, hidden_states: torch.Tensor,
+                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
+        logits = self.logits_processor(self.lm_head, hidden_states,
+                                       sampling_metadata)
+        return logits
+
+    def sample(
+        self,
+        logits: Optional[torch.Tensor],
+        sampling_metadata: SamplingMetadata,
+    ) -> Optional[SamplerOutput]:
+        next_tokens = self.sampler(logits, sampling_metadata)
+        return next_tokens
+
+    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            ("qkv_proj", "q_proj", "q"),
+            ("qkv_proj", "k_proj", "k"),
+            ("qkv_proj", "v_proj", "v"),
+            ("gate_up_proj", "gate_proj", 0),
+            ("gate_up_proj", "up_proj", 1),
+        ]
+
+        # Params for weights, fp8 weight scales, fp8 activation scales
+        # (param_name, weight_name, expert_id, shard_id)
+        expert_params_mapping = FusedMoE.make_expert_params_mapping(
+            ckpt_gate_proj_name="gate_proj",
+            ckpt_down_proj_name="down_proj",
+            ckpt_up_proj_name="up_proj",
+            num_experts=self.config.num_experts)
+
+        params_dict = dict(self.named_parameters())
+        for name, loaded_weight in weights:
+            if "rotary_emb.inv_freq" in name:
+                continue
+            for (param_name, weight_name, shard_id) in stacked_params_mapping:
+                # Skip non-stacked layers and experts (experts handled below).
+                if weight_name not in name:
+                    continue
+                # We have mlp.experts[0].gate_proj in the checkpoint.
+                # Since we handle the experts below in expert_params_mapping,
+                # we need to skip here BEFORE we update the name, otherwise
+                # name will be updated to mlp.experts[0].gate_up_proj, which
+                # will then be updated below in expert_params_mapping
+                # for mlp.experts[0].gate_gate_up_proj, which breaks load.
+                if "mlp.experts" in name:
+                    continue
+                name = name.replace(weight_name, param_name)
+                # Skip loading extra bias for GPTQ models.
+                if name.endswith(".bias") and name not in params_dict:
+                    continue
+                if name not in params_dict:
+                    continue
+
+                param = params_dict[name]
+                weight_loader = param.weight_loader
+                weight_loader(param, loaded_weight, shard_id)
+                break
+            else:
+                for mapping in expert_params_mapping:
+                    param_name, weight_name, expert_id, shard_id = mapping
+                    if weight_name not in name:
+                        continue
+                    name = name.replace(weight_name, param_name)
+                    param = params_dict[name]
+                    weight_loader = param.weight_loader
+                    weight_loader(param,
+                                  loaded_weight,
+                                  name,
+                                  shard_id=shard_id,
+                                  expert_id=expert_id)
+                    break
+                else:
+                    # Skip loading extra bias for GPTQ models.
+                    if name.endswith(".bias") and name not in params_dict:
+                        continue
+                    # Remapping the name of FP8 kv-scale.
+                    if name.endswith("kv_scale"):
+                        remapped_kv_scale_name = name.replace(
+                            ".kv_scale", ".attn.kv_scale")
+                        if remapped_kv_scale_name not in params_dict:
+                            print_warning_once(
+                                "Found kv scale in the checkpoint "
+                                f"(e.g. {name}), but not found the expected "
+                                f"name in the model "
+                                f"(e.g. {remapped_kv_scale_name}). "
+                                "kv-scale is not loaded.")
+                            continue
+                        else:
+                            name = remapped_kv_scale_name
+
+                    param = params_dict[name]
+                    weight_loader = getattr(param, "weight_loader",
+                                            default_weight_loader)
+                    weight_loader(param, loaded_weight)