Feat/searchmap preview

mteb/models/overview.py

@@ -66,6 +66,7 @@
     rerankers_monot5_based,
     ru_sentence_models,
     salesforce_models,
+    searchmap_models,
     sentence_transformers_models,
     siglip_models,
     sonar_models,
@@ -133,6 +134,7 @@
     rerankers_monot5_based,
     ru_sentence_models,
     salesforce_models,
+    searchmap_models,
     sentence_transformers_models,
     siglip_models,
     vista_models,

mteb/models/searchmap_models.py

@@ -0,0 +1,43 @@
+from __future__ import annotations
+
+from functools import partial
+
+from mteb.model_meta import ModelMeta, sentence_transformers_loader
+
+# Define task instructions with specific task names
+task_instructions = {
+    "Classification": "Generate a representation for this text that can be used for classification:",
+    "Clustering": "Generate a representation for this text that can be used for clustering:",
+    "Retrieval": "Generate a representation for this text that can be used for retrieval:",
+    "STS": "Generate a representation for this text that can be used for semantic similarity:",
+    "PairClassification": "Generate a representation for this text pair that can be used for classification:",
+    "Reranking": "Generate a representation for this text that can be used for reranking:",
+    "Summarization": "Generate a representation for this text that can be used for summarization:",
+}
+
+searchmap_preview = ModelMeta(
+    loader=partial(
+        sentence_transformers_loader,
+        model_name="VPLabs/SearchMap_Preview",
+        revision="69de17ef48278ed08ba1a4e65ead8179912b696e",
+        model_prompts=task_instructions,
+    ),
+    name="VPLabs/SearchMap_Preview",
+    revision="69de17ef48278ed08ba1a4e65ead8179912b696e",
+    languages=["eng_Latn"],
+    open_weights=True,
+    use_instructions=True,
+    release_date="2025-03-05",
+    n_parameters=435_000_000,
+    memory_usage_mb=1660,
+    embed_dim=4096,
+    license="mit",
+    max_tokens=8192,
+    reference="https://huggingface.co/VPLabs/SearchMap_Preview",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    public_training_code=None,
+    public_training_data=None,
+    training_datasets=None,
+    adapted_from="NovaSearch/stella_en_400M_v5",
+)