nextrec 0.4.20__py3-none-any.whl → 0.4.21__py3-none-any.whl

nextrec/models/multi_task/share_bottom.py

@@ -55,7 +55,7 @@ class ShareBottom(BaseModel):
 
     @property
     def default_task(self):
-        nums_task = getattr(self, "nums_task", None)
+        nums_task = self.nums_task if hasattr(self, "nums_task") else None
         if nums_task is not None and nums_task > 0:
             return ["binary"] * nums_task
         return ["binary"]
@@ -73,11 +73,10 @@ class ShareBottom(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | list[str | nn.Module] | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -103,7 +102,6 @@ class ShareBottom(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=resolved_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -125,7 +123,7 @@ class ShareBottom(BaseModel):
         # Calculate input dimension
         input_dim = self.embedding.input_dim
         # emb_dim_total = sum([f.embedding_dim for f in self.all_features if not isinstance(f, DenseFeature)])
-        # dense_input_dim = sum([getattr(f, "embedding_dim", 1) or 1 for f in dense_features])
+        # dense_input_dim = sum([(f.embedding_dim or 1) for f in dense_features])
         # input_dim = emb_dim_total + dense_input_dim
 
         # Shared bottom network

nextrec/models/ranking/afm.py

@@ -67,11 +67,10 @@ class AFM(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -88,7 +87,6 @@ class AFM(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/autoint.py

@@ -88,11 +88,10 @@ class AutoInt(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -102,7 +101,6 @@ class AutoInt(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/dcn.py

@@ -102,11 +102,10 @@ class DCN(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -123,7 +122,6 @@ class DCN(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -143,7 +141,10 @@ class DCN(BaseModel):
             ]
         )
         dense_input_dim = sum(
-            [getattr(f, "embedding_dim", 1) or 1 for f in dense_features]
+            [
+                (f.embedding_dim if f.embedding_dim is not None else 1) or 1
+                for f in dense_features
+            ]
         )
         input_dim = emb_dim_total + dense_input_dim
 

nextrec/models/ranking/dcn_v2.py

@@ -204,11 +204,10 @@ class DCNv2(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
         dense_features = dense_features or []
@@ -224,7 +223,6 @@ class DCNv2(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/deepfm.py

@@ -71,11 +71,10 @@ class DeepFM(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -93,7 +92,6 @@ class DeepFM(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -107,7 +105,7 @@ class DeepFM(BaseModel):
         self.embedding = EmbeddingLayer(features=self.deep_features)
         fm_emb_dim_total = sum([f.embedding_dim for f in self.fm_features])
         # deep_emb_dim_total = sum([f.embedding_dim for f in self.deep_features if not isinstance(f, DenseFeature)])
-        # dense_input_dim = sum([getattr(f, "embedding_dim", 1) or 1 for f in dense_features])
+        # dense_input_dim = sum([(f.embedding_dim or 1) for f in dense_features])
         mlp_input_dim = self.embedding.input_dim
         self.linear = LR(fm_emb_dim_total)
         self.fm = FM(reduce_sum=True)

nextrec/models/ranking/dien.py

@@ -220,11 +220,10 @@ class DIEN(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -243,7 +242,6 @@ class DIEN(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -342,7 +340,10 @@ class DIEN(BaseModel):
         mlp_input_dim += gru_hidden_size  # final interest state
         mlp_input_dim += sum([f.embedding_dim for f in self.other_sparse_features])
         mlp_input_dim += sum(
-            [getattr(f, "embedding_dim", 1) or 1 for f in dense_features]
+            [
+                (f.embedding_dim if f.embedding_dim is not None else 1) or 1
+                for f in dense_features
+            ]
         )
 
         self.mlp = MLP(input_dim=mlp_input_dim, **mlp_params)

nextrec/models/ranking/din.py

@@ -86,11 +86,10 @@ class DIN(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -109,7 +108,6 @@ class DIN(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -168,7 +166,10 @@ class DIN(BaseModel):
         mlp_input_dim += behavior_emb_dim  # attention pooled
         mlp_input_dim += sum([f.embedding_dim for f in self.other_sparse_features])
         mlp_input_dim += sum(
-            [getattr(f, "embedding_dim", 1) or 1 for f in dense_features]
+            [
+                (f.embedding_dim if f.embedding_dim is not None else 1) or 1
+                for f in dense_features
+            ]
         )
 
         # MLP for final prediction

nextrec/models/ranking/eulernet.py

@@ -211,11 +211,10 @@ class EulerNet(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -232,7 +231,6 @@ class EulerNet(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -245,7 +243,7 @@ class EulerNet(BaseModel):
 
         self.linear_features = dense_features + sparse_features + sequence_features
         self.interaction_features = (
-            [f for f in dense_features if getattr(f, "use_embedding", False)]
+            [f for f in dense_features if f.use_projection]
             + sparse_features
             + sequence_features
         )

nextrec/models/ranking/ffm.py

@@ -69,11 +69,10 @@ class FFM(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
         dense_features = dense_features or []
@@ -89,7 +88,6 @@ class FFM(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -169,7 +167,7 @@ class FFM(BaseModel):
         return f"{src_feature.embedding_name}__to__{target_field.name}"
 
     def build_embedding(self, feature: SparseFeature | SequenceFeature) -> nn.Embedding:
-        if getattr(feature, "pretrained_weight", None) is not None:
+        if feature.pretrained_weight is not None:
             weight = feature.pretrained_weight
             if weight is None:
                 raise ValueError(

nextrec/models/ranking/fibinet.py

@@ -80,11 +80,10 @@ class FiBiNET(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -102,7 +101,6 @@ class FiBiNET(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/fm.py

@@ -67,11 +67,10 @@ class FM(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -85,7 +84,6 @@ class FM(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/lr.py

@@ -66,11 +66,10 @@ class LR(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -87,7 +86,6 @@ class LR(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/masknet.py

@@ -179,11 +179,10 @@ class MaskNet(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg: float = 1e-6,
-        dense_l1_reg: float = 1e-5,
-        embedding_l2_reg: float = 1e-5,
-        dense_l2_reg: float = 1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
         dense_features = dense_features or []
@@ -199,7 +198,6 @@ class MaskNet(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -220,16 +218,17 @@ class MaskNet(BaseModel):
         ), "MaskNet requires at least one feature for masking."
         self.embedding = EmbeddingLayer(features=self.mask_features)
         self.num_fields = len(self.mask_features)
-        self.embedding_dim = getattr(self.mask_features[0], "embedding_dim", None)
+        self.embedding_dim = self.mask_features[0].embedding_dim
         assert (
             self.embedding_dim is not None
         ), "MaskNet requires mask_features to have 'embedding_dim' defined."
 
         for f in self.mask_features:
-            edim = getattr(f, "embedding_dim", None)
+            edim = f.embedding_dim
             if edim is None or edim != self.embedding_dim:
+                feat_name = f.name if hasattr(f, "name") else type(f)
                 raise ValueError(
-                    f"MaskNet expects identical embedding_dim across all mask_features, but got {edim} for feature {getattr(f, 'name', type(f))}."
+                    f"MaskNet expects identical embedding_dim across all mask_features, but got {edim} for feature {feat_name}."
                 )
 
         self.v_emb_dim = self.num_fields * self.embedding_dim

nextrec/models/ranking/pnn.py

@@ -67,11 +67,10 @@ class PNN(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -91,7 +90,6 @@ class PNN(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,

nextrec/models/ranking/widedeep.py

@@ -67,11 +67,10 @@ class WideDeep(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -87,7 +86,6 @@ class WideDeep(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -112,7 +110,7 @@ class WideDeep(BaseModel):
         # Deep part: MLP
         input_dim = self.embedding.input_dim
         # deep_emb_dim_total = sum([f.embedding_dim for f in self.deep_features if not isinstance(f, DenseFeature)])
-        # dense_input_dim = sum([getattr(f, "embedding_dim", 1) or 1 for f in dense_features])
+        # dense_input_dim = sum([(f.embedding_dim or 1) for f in dense_features])
         self.mlp = MLP(input_dim=input_dim, **mlp_params)
         self.prediction_layer = TaskHead(task_type=self.task)
         # Register regularization weights

nextrec/models/ranking/xdeepfm.py

@@ -127,11 +127,10 @@ class xDeepFM(BaseModel):
         optimizer_params: dict | None = None,
         loss: str | nn.Module | None = "bce",
         loss_params: dict | list[dict] | None = None,
-        device: str = "cpu",
-        embedding_l1_reg=1e-6,
-        dense_l1_reg=1e-5,
-        embedding_l2_reg=1e-5,
-        dense_l2_reg=1e-4,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         **kwargs,
     ):
 
@@ -148,7 +147,6 @@ class xDeepFM(BaseModel):
             sequence_features=sequence_features,
             target=target,
             task=task or self.default_task,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
@@ -184,7 +182,10 @@ class xDeepFM(BaseModel):
             ]
         )
         dense_input_dim = sum(
-            [getattr(f, "embedding_dim", 1) or 1 for f in dense_features]
+            [
+                (f.embedding_dim if f.embedding_dim is not None else 1) or 1
+                for f in dense_features
+            ]
         )
         self.mlp = MLP(input_dim=deep_emb_dim_total + dense_input_dim, **mlp_params)
         self.prediction_layer = TaskHead(task_type=self.task)

nextrec/models/retrieval/dssm.py

@@ -50,12 +50,10 @@ class DSSM(BaseMatchModel):
         num_negative_samples: int = 4,
         temperature: float = 1.0,
         similarity_metric: Literal["dot", "cosine", "euclidean"] = "cosine",
-        device: str = "cpu",
-        embedding_l1_reg: float = 0.0,
-        dense_l1_reg: float = 0.0,
-        embedding_l2_reg: float = 0.0,
-        dense_l2_reg: float = 0.0,
-        early_stop_patience: int = 20,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         optimizer: str | torch.optim.Optimizer = "adam",
         optimizer_params: dict | None = None,
         scheduler: (
@@ -81,12 +79,10 @@ class DSSM(BaseMatchModel):
             num_negative_samples=num_negative_samples,
             temperature=temperature,
             similarity_metric=similarity_metric,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
             dense_l2_reg=dense_l2_reg,
-            early_stop_patience=early_stop_patience,
             **kwargs,
         )
 
@@ -175,8 +171,6 @@ class DSSM(BaseMatchModel):
             loss_params=loss_params,
         )
 
-        self.to(device)
-
     def user_tower(self, user_input: dict) -> torch.Tensor:
         """
         User tower encodes user features into embeddings.

nextrec/models/retrieval/dssm_v2.py

@@ -46,12 +46,10 @@ class DSSM_v2(BaseMatchModel):
         num_negative_samples: int = 4,
         temperature: float = 1.0,
         similarity_metric: Literal["dot", "cosine", "euclidean"] = "dot",
-        device: str = "cpu",
         embedding_l1_reg: float = 0.0,
         dense_l1_reg: float = 0.0,
         embedding_l2_reg: float = 0.0,
         dense_l2_reg: float = 0.0,
-        early_stop_patience: int = 20,
         optimizer: str | torch.optim.Optimizer = "adam",
         optimizer_params: dict | None = None,
         scheduler: (
@@ -77,12 +75,10 @@ class DSSM_v2(BaseMatchModel):
             num_negative_samples=num_negative_samples,
             temperature=temperature,
             similarity_metric=similarity_metric,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
             dense_l2_reg=dense_l2_reg,
-            early_stop_patience=early_stop_patience,
             **kwargs,
         )
 
@@ -167,8 +163,6 @@ class DSSM_v2(BaseMatchModel):
             loss_params=loss_params,
         )
 
-        self.to(device)
-
     def user_tower(self, user_input: dict) -> torch.Tensor:
         """User tower"""
         all_user_features = (

nextrec/models/retrieval/mind.py

@@ -202,12 +202,10 @@ class MIND(BaseMatchModel):
         num_negative_samples: int = 100,
         temperature: float = 1.0,
         similarity_metric: Literal["dot", "cosine", "euclidean"] = "dot",
-        device: str = "cpu",
-        embedding_l1_reg: float = 0.0,
-        dense_l1_reg: float = 0.0,
-        embedding_l2_reg: float = 0.0,
-        dense_l2_reg: float = 0.0,
-        early_stop_patience: int = 20,
+        embedding_l1_reg=0.0,
+        dense_l1_reg=0.0,
+        embedding_l2_reg=0.0,
+        dense_l2_reg=0.0,
         optimizer: str | torch.optim.Optimizer = "adam",
         optimizer_params: dict | None = None,
         scheduler: (
@@ -233,12 +231,10 @@ class MIND(BaseMatchModel):
             num_negative_samples=num_negative_samples,
             temperature=temperature,
             similarity_metric=similarity_metric,
-            device=device,
             embedding_l1_reg=embedding_l1_reg,
             dense_l1_reg=dense_l1_reg,
             embedding_l2_reg=embedding_l2_reg,
             dense_l2_reg=dense_l2_reg,
-            early_stop_patience=early_stop_patience,
             **kwargs,
         )
 
@@ -335,8 +331,6 @@ class MIND(BaseMatchModel):
             loss_params=loss_params,
         )
 
-        self.to(device)
-
     def user_tower(self, user_input: dict) -> torch.Tensor:
         """
         User tower with multi-interest extraction