diff --git a/docs/encoders.md b/docs/encoders.md
index ef791281..3ea87a7d 100644
--- a/docs/encoders.md
+++ b/docs/encoders.md
@@ -1,20 +1,32 @@
 # Encoders
 
+!!! warning
+    We now recommend that you do NOT pre-load `SentenceTransformer` encoder models, but rather pass them by name to the topic models.
+    This allows the model not to store the encoder model on disk, and load it when loading the model.
+    This can lead to extreme reductions in disk space usage.
+    For example, instead of writing:
+    ```python
+    model = SensTopic(
+        encoder=SentenceTransformer("intfloat/multilingual-e5-large-instruct", default_prompt_name="query")
+    )
+    ```
+    Write:
+    ```python
+    model = SensTopic(
+        encoder="intfloat/multilingual-e5-large-instruct",
+        trf_kwargs=dict(default_prompt_name="query")
+    )
+    ```
+
+
 Turftopic by default encodes documents using sentence transformers.
 You can always change the encoder model either by passing the name of a sentence transformer from the Huggingface Hub to a model, or by passing a `SentenceTransformer` instance.
 
 Here's an example of building a multilingual topic model by using multilingual embeddings:
 
 ```python
-from sentence_transformers import SentenceTransformer
 from turftopic import GMM
 
-trf = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")
-
-model = GMM(10, encoder=trf)
-
-# or
-
 model = GMM(10, encoder="paraphrase-multilingual-MiniLM-L12-v2")
 ```
 
@@ -35,31 +47,36 @@ In this case, documents will serve as the queries and words as the passages:
 from turftopic import KeyNMF
 from sentence_transformers import SentenceTransformer
 
-encoder = SentenceTransformer(
-    "intfloat/multilingual-e5-large-instruct",
-    prompts={
-        "query": "Instruct: Retrieve relevant keywords from the given document. Query: "
-        "passage": "Passage: "
-    },
-    # Make sure to set default prompt to query!
-    default_prompt_name="query",
+model = KeyNMF(
+    10,
+    encoder="intfloat/multilingual-e5-large-instruct",
+    # These are the arguments that get applied when loading the encoder.
+    trf_kwargs=dict(
+        prompts={
+            "query": "Instruct: Retrieve relevant keywords from the given document. Query: "
+            "passage": "Passage: "
+        },
+        # Make sure to set default prompt to query!
+        default_prompt_name="query",
+    )
 )
-model = KeyNMF(10, encoder=encoder)
 ```
 
 And a regular, asymmetric example:
 
 ```python
-encoder = SentenceTransformer(
-    "intfloat/e5-large-v2",
-    prompts={
-        "query": "query: "
-        "passage": "passage: "
-    },
-    # Make sure to set default prompt to query!
-    default_prompt_name="query",
+model = KeyNMF(
+    10,
+    encoder="intfloat/e5-large-v2",
+    trf_kwargs=dict(
+        prompts={
+            "query": "query: "
+            "passage": "passage: "
+        },
+        # Make sure to set default prompt to query!
+        default_prompt_name="query",
+    )
 )
-model = KeyNMF(10, encoder=encoder)
 ```
 
 ## Performance tips
@@ -75,9 +92,8 @@ pip install sentence-transformers[onnx, onnx-gpu]
 from turftopic import SemanticSignalSeparation
 from sentence_transformers import SentenceTransformer
 
-encoder = SentenceTransformer("all-MiniLM-L6-v2", backend="onnx")
 
-model = SemanticSignalSeparation(10, encoder=encoder)
+model = SemanticSignalSeparation(10, encoder="all-MiniLM-L6-v2", trf_kwargs=dict(backend="onnx"))
 ```
 
 ## External Embeddings
diff --git a/docs/persistence.md b/docs/persistence.md
index 94825570..12dc0315 100644
--- a/docs/persistence.md
+++ b/docs/persistence.md
@@ -3,6 +3,24 @@
 ## Model persistence
 All models in Turftopic can be serialized and saved to disk, or published to the HuggingFace Hub.
 
+!!! warning
+    We now recommend that you do NOT pre-load `SentenceTransformer` encoder models, but rather pass them by name to the topic models.
+    This allows the model not to store the encoder model on disk, and load it when loading the model.
+    This can lead to extreme reductions in disk space usage.
+    For example, instead of writing:
+    ```python
+    model = SensTopic(
+        encoder=SentenceTransformer("intfloat/multilingual-e5-large-instruct", default_prompt_name="query")
+    )
+    ```
+    Write:
+    ```python
+    model = SensTopic(
+        encoder="intfloat/multilingual-e5-large-instruct",
+        trf_kwargs=dict(default_prompt_name="query")
+    )
+    ```
+
 ### Saving locally
 
 Turftopic models can now be saved to disk using the `to_disk()` method of models:
diff --git a/pyproject.toml b/pyproject.toml
index eded9c19..db3cc10c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -9,7 +9,7 @@ profile = "black"
 
 [project]
 name = "turftopic"
-version = "0.26.1"
+version = "0.27.0"
 description = "Topic modeling with contextual representations from sentence transformers."
 authors = [
    { name = "Márton Kardos <power.up1163@gmail.com>", email = "martonkardos@cas.au.dk" }
diff --git a/turftopic/base.py b/turftopic/base.py
index 6606c650..d74f6987 100644
--- a/turftopic/base.py
+++ b/turftopic/base.py
@@ -22,6 +22,18 @@
 class ContextualModel(BaseEstimator, TransformerMixin, TopicContainer):
     """Base class for contextual topic models in Turftopic."""
 
+    def load_encoder(self):
+        if isinstance(self.encoder, str):
+            if self.trf_kwargs is None:
+                trf_kwargs = dict()
+            else:
+                trf_kwargs = self.trf_kwargs
+            self.encoder_ = SentenceTransformer(self.encoder, **trf_kwargs)
+            self._encoder_preloaded = False
+        else:
+            self.encoder_ = self.encoder
+            self._encoder_preloaded = True
+
     @property
     def has_negative_side(self) -> bool:
         return False
@@ -41,7 +53,11 @@ def encode_documents(self, raw_documents: Iterable[str]) -> np.ndarray:
         """
         if not hasattr(self.encoder_, "encode"):
             return self.encoder.get_text_embeddings(list(raw_documents))
-        return self.encoder_.encode(list(raw_documents))
+        if getattr(self, "encode_kwargs", None) is None:
+            encode_kwargs = dict()
+        else:
+            encode_kwargs = self.encode_kwargs
+        return self.encoder_.encode(list(raw_documents), **encode_kwargs)
 
     @abstractmethod
     def fit_transform(
@@ -173,7 +189,13 @@ def to_disk(self, out_dir: Union[Path, str]):
         package_versions = get_package_versions()
         with out_dir.joinpath("package_versions.json").open("w") as ver_file:
             ver_file.write(json.dumps(package_versions))
-        joblib.dump(self, out_dir.joinpath("model.joblib"))
+        if getattr(self, "_encoder_preloaded", True):
+            joblib.dump(self, out_dir.joinpath("model.joblib"))
+        else:
+            encoder_ = self.encoder_
+            delattr(self, "encoder_")
+            joblib.dump(self, out_dir.joinpath("model.joblib"))
+            self.encoder_ = encoder_
 
     def push_to_hub(self, repo_id: str):
         """Uploads model to HuggingFace Hub
diff --git a/turftopic/models/cluster.py b/turftopic/models/cluster.py
index feb59988..ca828fba 100644
--- a/turftopic/models/cluster.py
+++ b/turftopic/models/cluster.py
@@ -180,6 +180,10 @@ class ClusteringTopicModel(
         if 'centroid' the centroid vectors of clusters will be used as topic vectors (Top2Vec).
     random_state: int, default None
         Random state to use so that results are exactly reproducible.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -196,9 +200,13 @@ def __init__(
         reduction_distance_metric: DistanceMetric = "cosine",
         reduction_topic_representation: TopicRepresentation = "component",
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.encoder = encoder
         self.random_state = random_state
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
         if feature_importance not in VALID_WORD_IMPORTANCE:
             raise ValueError(
                 f"feature_importance must be one of {VALID_WORD_IMPORTANCE} got {feature_importance} instead."
@@ -217,10 +225,7 @@ def __init__(
             )
         if isinstance(encoder, int):
             raise TypeError(integer_message)
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.load_encoder()
         self.validate_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
@@ -766,6 +771,8 @@ def __init__(
         reduction_distance_metric: DistanceMetric = "cosine",
         reduction_topic_representation: TopicRepresentation = "component",
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         if dimensionality_reduction is None:
             try:
@@ -798,6 +805,8 @@ def __init__(
             reduction_method=reduction_method,
             reduction_distance_metric=reduction_distance_metric,
             reduction_topic_representation=reduction_topic_representation,
+            trf_kwargs=trf_kwargs,
+            encode_kwargs=encode_kwargs,
         )
 
 
@@ -834,6 +843,8 @@ def __init__(
         reduction_distance_metric: DistanceMetric = "cosine",
         reduction_topic_representation: TopicRepresentation = "centroid",
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         if dimensionality_reduction is None:
             try:
@@ -866,6 +877,8 @@ def __init__(
             reduction_method=reduction_method,
             reduction_distance_metric=reduction_distance_metric,
             reduction_topic_representation=reduction_topic_representation,
+            trf_kwargs=trf_kwargs,
+            encode_kwargs=encode_kwargs,
         )
 
 
@@ -911,6 +924,8 @@ def __init__(
         step_size: Optional[int] = 40,
         pooling: Optional[Callable] = np.nanmean,
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         if dimensionality_reduction is None:
             try:
@@ -956,6 +971,8 @@ def __init__(
             reduction_method=reduction_method,
             reduction_distance_metric=reduction_distance_metric,
             reduction_topic_representation=reduction_topic_representation,
+            trf_kwargs=trf_kwargs,
+            encode_kwargs=encode_kwargs,
         )
         super().__init__(
             model,
diff --git a/turftopic/models/ctm.py b/turftopic/models/ctm.py
index e432f4b5..046569e7 100644
--- a/turftopic/models/ctm.py
+++ b/turftopic/models/ctm.py
@@ -139,6 +139,10 @@ class AutoEncodingTopicModel(ContextualModel, MultimodalModel):
         Number of epochs to run during training.
     random_state: int, default None
         Random state to use so that results are exactly reproducible.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -155,14 +159,15 @@ def __init__(
         learning_rate: float = 1e-2,
         n_epochs: int = 50,
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.n_components = n_components
         self.random_state = random_state
         self.encoder = encoder
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
+        self.load_encoder()
         self.validate_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
diff --git a/turftopic/models/cvp.py b/turftopic/models/cvp.py
index 8b9b64c1..37c985d2 100644
--- a/turftopic/models/cvp.py
+++ b/turftopic/models/cvp.py
@@ -10,14 +10,14 @@
 from sentence_transformers import SentenceTransformer
 from sklearn.base import BaseEstimator, TransformerMixin
 
-from turftopic.base import Encoder
+from turftopic.base import ContextualModel, Encoder
 from turftopic.encoders.multimodal import MultimodalEncoder
 from turftopic.serialization import create_readme, get_package_versions
 
 Seeds = tuple[list[str], list[str]]
 
 
-class ConceptVectorProjection(BaseEstimator, TransformerMixin):
+class ConceptVectorProjection(ContextualModel):
     """Concept Vector Projection model from [Lyngbæk et al. (2025)](https://doi.org/10.63744/nVu1Zq5gRkuD)
     Can be used to project document embeddings onto a difference projection vector between positive and negative seed phrases.
     The primary use case is sentiment analysis, and continuous sentiment scores,
@@ -34,6 +34,10 @@ class ConceptVectorProjection(BaseEstimator, TransformerMixin):
     encoder: str or SentenceTransformer
         Model to produce document representations, paraphrase-multilingual-mpnet-base-v2 is the default
         per Lyngbæk et al. (2025).
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -42,6 +46,8 @@ def __init__(
         encoder: Union[
             Encoder, str, MultimodalEncoder
         ] = "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.seeds = seeds
         if isinstance(seeds, OrderedDict):
@@ -55,15 +61,14 @@ def __init__(
         else:
             self._seeds = OrderedDict(seeds)
         self.encoder = encoder
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
+        self.load_encoder()
         self.classes_ = np.array([name for name in self._seeds])
         self.concept_matrix_ = []
         for _, (positive, negative) in self._seeds.items():
-            positive_emb = self.encoder_.encode(list(positive))
-            negative_emb = self.encoder_.encode(list(negative))
+            positive_emb = self.encode_documents(list(positive))
+            negative_emb = self.encode_documents(list(negative))
             cv = np.mean(positive_emb, axis=0) - np.mean(negative_emb, axis=0)
             self.concept_matrix_.append(cv / np.linalg.norm(cv))
         self.concept_matrix_ = np.stack(self.concept_matrix_)
@@ -92,7 +97,7 @@ def fit_transform(self, raw_documents=None, y=None, embeddings=None):
                 "Either embeddings or raw_documents has to be passed, both are None."
             )
         if embeddings is None:
-            embeddings = self.encoder_.encode(list(raw_documents))
+            embeddings = self.encode_documents(list(raw_documents))
         return embeddings @ self.concept_matrix_.T
 
     def transform(self, raw_documents=None, embeddings=None):
@@ -111,39 +116,3 @@ def transform(self, raw_documents=None, embeddings=None):
             Prevalance of each concept in each document.
         """
         return self.fit_transform(raw_documents, embeddings=embeddings)
-
-    def to_disk(self, out_dir: Union[Path, str]):
-        """Persists model to directory on your machine.
-
-        Parameters
-        ----------
-        out_dir: Path | str
-            Directory to save the model to.
-        """
-        out_dir = Path(out_dir)
-        out_dir.mkdir(exist_ok=True)
-        package_versions = get_package_versions()
-        with out_dir.joinpath("package_versions.json").open("w") as ver_file:
-            ver_file.write(json.dumps(package_versions))
-        joblib.dump(self, out_dir.joinpath("model.joblib"))
-
-    def push_to_hub(self, repo_id: str):
-        """Uploads model to HuggingFace Hub
-
-        Parameters
-        ----------
-        repo_id: str
-            Repository to upload the model to.
-        """
-        api = HfApi()
-        api.create_repo(repo_id, exist_ok=True)
-        with tempfile.TemporaryDirectory() as tmp_dir:
-            readme_path = Path(tmp_dir).joinpath("README.md")
-            with readme_path.open("w") as readme_file:
-                readme_file.write(create_readme(self, repo_id))
-            self.to_disk(tmp_dir)
-            api.upload_folder(
-                folder_path=tmp_dir,
-                repo_id=repo_id,
-                repo_type="model",
-            )
diff --git a/turftopic/models/decomp.py b/turftopic/models/decomp.py
index d2a3facd..00d150a4 100644
--- a/turftopic/models/decomp.py
+++ b/turftopic/models/decomp.py
@@ -67,6 +67,10 @@ class SemanticSignalSeparation(
         or their combination ('combined') should determine the word's importance for a topic.
     random_state: int, default None
         Random state to use so that results are exactly reproducible.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -82,14 +86,15 @@ def __init__(
             "axial", "angular", "combined"
         ] = "combined",
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.n_components = n_components
         self.encoder = encoder
         self.feature_importance = feature_importance
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
+        self.load_encoder()
         self.validate_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
diff --git a/turftopic/models/fastopic.py b/turftopic/models/fastopic.py
index 1642964b..bec0bc3f 100644
--- a/turftopic/models/fastopic.py
+++ b/turftopic/models/fastopic.py
@@ -59,6 +59,10 @@ class FASTopic(ContextualModel):
         Learning rate for the ADAM optimizer.
     device: str, default "cpu"
         Device to run the model on. Defaults to CPU.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -76,14 +80,15 @@ def __init__(
         n_epochs: int = 200,
         learning_rate: float = 0.002,
         device: str = "cpu",
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.n_components = n_components
         self.encoder = encoder
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
         self.random_state = random_state
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.load_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
         else:
@@ -153,7 +158,7 @@ def fit_transform(
         with console.status("Fitting model") as status:
             if embeddings is None:
                 status.update("Encoding documents")
-                embeddings = self.encoder_.encode(raw_documents)
+                embeddings = self.encode_documents(raw_documents)
                 console.log("Documents encoded.")
             self.train_doc_embeddings = embeddings
             status.update("Extracting terms.")
@@ -188,7 +193,7 @@ def transform(
             Document-topic matrix.
         """
         if embeddings is None:
-            embeddings = self.encoder_.encode(raw_documents)
+            embeddings = self.encode_documents(raw_documents)
         with torch.no_grad():
             self.model.eval()
             theta = self.model.get_theta(
diff --git a/turftopic/models/gmm.py b/turftopic/models/gmm.py
index cf2b7c53..04d5210b 100644
--- a/turftopic/models/gmm.py
+++ b/turftopic/models/gmm.py
@@ -91,6 +91,10 @@ class GMM(ContextualModel, DynamicTopicModel, MultimodalModel):
             not embedding-based ones.*
         random_state: int, default None
             Random state to use so that results are exactly reproducible.
+        trf_kwargs: dict, default None
+            Keyword arguments to apply when loading the Encoder model.
+        encode_kwargs: dict, default None
+            Keyword arguments to apply encoding documents with the encoder.
 
         Attributes
         ----------
@@ -110,16 +114,17 @@ def __init__(
         weight_prior: Literal["dirichlet", "dirichlet_process", None] = None,
         gamma: Optional[float] = None,
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.n_components = n_components
         self.encoder = encoder
+        self.encode_kwargs = encode_kwargs
+        self.trf_kwargs = trf_kwargs
         self.weight_prior = weight_prior
         self.gamma = gamma
         self.random_state = random_state
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.load_encoder()
         self.validate_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
diff --git a/turftopic/models/keynmf.py b/turftopic/models/keynmf.py
index c9618a7b..f00562a6 100644
--- a/turftopic/models/keynmf.py
+++ b/turftopic/models/keynmf.py
@@ -69,6 +69,10 @@ class KeyNMF(ContextualModel, DynamicTopicModel, MultimodalModel):
         This is useful when you have a corpus containing multiple languages.
     term_match_threshold: float, default 0.9
         Cosine similarity threshold for matching terms across languages.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -85,6 +89,8 @@ def __init__(
         seed_exponent: float = 2.0,
         cross_lingual: bool = False,
         term_match_threshold: float = 0.9,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.random_state = random_state
         self.n_components = n_components
@@ -92,6 +98,8 @@ def __init__(
         self.seed_exponent = seed_exponent
         self.metric = metric
         self.encoder = encoder
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
         self._has_custom_vectorizer = vectorizer is not None
         if isinstance(encoder, str):
             if (
@@ -101,9 +109,7 @@ def __init__(
                     "all-MiniLM is incompatible with cross-lingual transfer, using paraphrase-multilingual-MiniLM-L12-v2."
                 )
                 encoder = "paraphrase-multilingual-MiniLM-L12-v2"
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.load_encoder()
         self.validate_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
@@ -121,7 +127,7 @@ def __init__(
         self.seed_phrase = seed_phrase
         self.seed_embedding = None
         if self.seed_phrase is not None:
-            self.seed_embedding = self.encoder_.encode([self.seed_phrase])[0]
+            self.seed_embedding = self.encode_documents([self.seed_phrase])[0]
         self.cross_lingual = cross_lingual
         self.term_match_threshold = term_match_threshold
 
diff --git a/turftopic/models/senstopic.py b/turftopic/models/senstopic.py
index 3a992464..016543dc 100644
--- a/turftopic/models/senstopic.py
+++ b/turftopic/models/senstopic.py
@@ -89,6 +89,10 @@ class SensTopic(ContextualModel, DynamicTopicModel, MultimodalModel):
         L1 penalty applied to document-topic proportions.
         Higher values push the model to assign fewer topics to a single document,
         while lower values will distribute topics across documents.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
     """
 
     def __init__(
@@ -104,14 +108,15 @@ def __init__(
         ] = "combined",
         random_state: Optional[int] = None,
         sparsity: float = 1,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         self.n_components = n_components
         self.encoder = encoder
+        self.trf_kwargs = trf_kwargs
+        self.encode_kwargs = encode_kwargs
         self.feature_importance = feature_importance
-        if isinstance(encoder, str):
-            self.encoder_ = SentenceTransformer(encoder)
-        else:
-            self.encoder_ = encoder
+        self.load_encoder()
         self.validate_encoder()
         if vectorizer is None:
             self.vectorizer = default_vectorizer()
@@ -346,7 +351,7 @@ def partial_fit(
 
     def transform(self, raw_documents, embeddings=None):
         if embeddings is None:
-            embeddings = self.encoder_.encode(raw_documents)
+            embeddings = self.encode_documents(raw_documents)
         return self.decomposition.transform(embeddings)
 
     def fit_transform_multimodal(
diff --git a/turftopic/models/topeax.py b/turftopic/models/topeax.py
index 52f05742..e3c7cc00 100644
--- a/turftopic/models/topeax.py
+++ b/turftopic/models/topeax.py
@@ -141,6 +141,10 @@ class Topeax(GMM):
         Number of neighbours to take into account when running TSNE.
     random_state: int, default None
         Random state to use so that results are exactly reproducible.
+    trf_kwargs: dict, default None
+        Keyword arguments to apply when loading the Encoder model.
+    encode_kwargs: dict, default None
+        Keyword arguments to apply encoding documents with the encoder.
 
     """
 
@@ -152,6 +156,8 @@ def __init__(
         vectorizer: Optional[CountVectorizer] = None,
         perplexity: int = 50,
         random_state: Optional[int] = None,
+        trf_kwargs=None,
+        encode_kwargs=None,
     ):
         dimensionality_reduction = TSNE(
             2,
@@ -166,6 +172,8 @@ def __init__(
             vectorizer=vectorizer,
             dimensionality_reduction=dimensionality_reduction,
             random_state=random_state,
+            trf_kwargs=trf_kwargs,
+            encode_kwargs=encode_kwargs,
         )
 
     def estimate_components(
diff --git a/turftopic/serialization.py b/turftopic/serialization.py
index 8336a4ed..5ffccf90 100644
--- a/turftopic/serialization.py
+++ b/turftopic/serialization.py
@@ -110,7 +110,12 @@ def load_model(repo_id_or_path: Union[str, Path]):
             remote_versions = json.loads(ver_file.read())
         validate_package_versions(remote_versions)
         model = joblib.load(path.joinpath("model.joblib"))
-        return model
     else:
         in_dir = snapshot_download(repo_id=repo_id_or_path)
-        return load_model(in_dir)
+        model = load_model(in_dir)
+    if getattr(model, "encoder_", None) is None:
+        print(
+            "Model does not come with prepackaged encoder_ attribute, loading it."
+        )
+        model.load_encoder()
+    return model