diff --git a/awq/models/__init__.py b/awq/models/__init__.py
index e2dfb0d7..4ea17bfb 100644
--- a/awq/models/__init__.py
+++ b/awq/models/__init__.py
@@ -20,5 +20,6 @@
 from .llava_next import LlavaNextAWQForCausalLM
 from .phi3 import Phi3AWQForCausalLM
 from .cohere import CohereAWQForCausalLM
+from .deepseek import DeepseekAWQForCausalLM
 from .deepseek_v2 import DeepseekV2AWQForCausalLM
 from .minicpm import MiniCPMAWQForCausalLM
diff --git a/awq/models/auto.py b/awq/models/auto.py
index 92a69b63..b9afb34e 100644
--- a/awq/models/auto.py
+++ b/awq/models/auto.py
@@ -30,6 +30,7 @@
     "llava_next": LlavaNextAWQForCausalLM,
     "phi3": Phi3AWQForCausalLM,
     "cohere": CohereAWQForCausalLM,
+    "deepseek": DeepseekAWQForCausalLM,
     "deepseek_v2": DeepseekV2AWQForCausalLM,
     "minicpm": MiniCPMAWQForCausalLM,
 }
diff --git a/awq/models/base.py b/awq/models/base.py
index da7117d8..d1ab8134 100644
--- a/awq/models/base.py
+++ b/awq/models/base.py
@@ -81,6 +81,7 @@
     "llava_next": "AutoModelForVision2Seq",
     "phi3": "AutoModelForCausalLM",
     "cohere": "AutoModelForCausalLM",
+    "deepseek": "AutoModelForCausalLM",
     "deepseek_v2": "AutoModelForCausalLM",
     "minicpm": "AutoModelForCausalLM",
 }
diff --git a/awq/models/deepseek.py b/awq/models/deepseek.py
new file mode 100644
index 00000000..f5f1c4a8
--- /dev/null
+++ b/awq/models/deepseek.py
@@ -0,0 +1,103 @@
+import tqdm
+from typing import List, Tuple
+from .base import BaseAWQForCausalLM
+
+
+class DeepseekAWQForCausalLM(BaseAWQForCausalLM):
+    layer_type = "DeepseekDecoderLayer"
+    max_seq_len_key = "max_position_embeddings"
+
+    @staticmethod
+    def get_model_layers(model):
+        return model.model.layers
+
+    @staticmethod
+    def get_act_for_scaling(module):
+        return dict(is_scalable=False)
+
+    @staticmethod
+    def move_embed(model, device: str):
+        model.model.embed_tokens = model.model.embed_tokens.to(device)
+
+    @staticmethod
+    def get_layers_for_scaling(
+        module, input_feat, module_kwargs
+    ):
+        layers = []
+        
+        # attention input
+        layers.append(
+            dict(
+                prev_op=module.input_layernorm,
+                layers=[
+                    module.self_attn.q_proj,
+                    module.self_attn.k_proj,
+                    module.self_attn.v_proj,
+                ],
+                inp=input_feat["self_attn.q_proj"],
+                module2inspect=module.self_attn,
+                kwargs=module_kwargs,
+            )
+        )
+        
+        if module.self_attn.v_proj.weight.shape == module.self_attn.o_proj.weight.shape:
+            layers.append(
+                dict(
+                    prev_op=module.self_attn.v_proj,
+                    layers=[module.self_attn.o_proj],
+                    inp=input_feat["self_attn.o_proj"],
+                )
+            )
+        
+        if hasattr(module.mlp, "gate"):
+            # linear in
+            layers.append(
+                dict(
+                    prev_op=module.post_attention_layernorm,
+                    layers=[
+                        w
+                        for expert in module.mlp.experts
+                        for w in [expert.gate_proj, expert.up_proj]
+                    ] + [module.mlp.shared_experts.gate_proj, module.mlp.shared_experts.up_proj],
+                    inp=input_feat["mlp"],
+                    module2inspect=module.mlp,
+                )
+            )
+
+            # linear out
+            for i, expert in enumerate(module.mlp.experts):
+                layers.append(
+                    dict(
+                        prev_op=expert.up_proj,
+                        layers=[expert.down_proj],
+                        inp=input_feat[f"mlp.experts.{i}.down_proj"],
+                    )
+                )
+            layers.append(
+                dict(
+                    prev_op=module.mlp.shared_experts.up_proj,
+                    layers=[module.mlp.shared_experts.down_proj],
+                    inp=input_feat[f"mlp.shared_experts.down_proj"],
+                )
+            )
+        else:
+            # linear 1
+            layers.append(
+                dict(
+                    prev_op=module.post_attention_layernorm,
+                    layers=[module.mlp.gate_proj, module.mlp.up_proj],
+                    inp=input_feat["mlp.gate_proj"],
+                    module2inspect=module.mlp,
+                )
+            )
+
+            # linear 2
+            layers.append(
+                dict(
+                    prev_op=module.mlp.up_proj,
+                    layers=[module.mlp.down_proj],
+                    inp=input_feat["mlp.down_proj"],
+                )
+            )
+
+        return layers
diff --git a/awq/quantize/quantizer.py b/awq/quantize/quantizer.py
index cd9fb0dd..f58288fa 100644
--- a/awq/quantize/quantizer.py
+++ b/awq/quantize/quantizer.py
@@ -610,7 +610,7 @@ def cache_input_hook(m, x, y, name, feat_dict):
                 "block_sparse_moe": layer.block_sparse_moe,
             }
 
-        if self.awq_model.model_type == "deepseek_v2":
+        if self.awq_model.model_type in  ["deepseek_v2", "deepseek"]:
             named_linears = {
                 **named_linears,
                 "mlp": layer.mlp,