Add IBM Granite Embedding Models (#1613)

* add IBM granite embedding models
* lint formatting
* add adapted_from and superseded_by to ModelMeta

mteb/models/ibm_granite_models.py

@@ -0,0 +1,114 @@
+from __future__ import annotations
+
+from functools import partial
+
+from mteb.model_meta import ModelMeta, sentence_transformers_loader
+
+GRANITE_LANGUAGES = [
+    "ara_Latn",
+    "ces_Latn",
+    "deu_Latn",
+    "eng_Latn",
+    "spa_Latn",
+    "fra_Latn",
+    "ita_Latn",
+    "jpn_Latn",
+    "kor_Latn",
+    "nld_Latn",
+    "por_Latn",
+    "zho_Hant",
+    "zho_Hans",
+]
+
+
+granite_107m_multilingual = ModelMeta(
+    loader=partial(  # type: ignore
+        sentence_transformers_loader,
+        model_name="ibm-granite/granite-embedding-107m-multilingual",
+        revision="47db56afe692f731540413c67dd818ff492277e7",
+    ),
+    name="ibm-granite/granite-embedding-107m-multilingual",
+    languages=GRANITE_LANGUAGES,
+    open_weights=True,
+    revision="47db56afe692f731540413c67dd818ff492277e7",
+    release_date="2024-12-18",
+    n_parameters=107_000_000,
+    memory_usage=None,
+    embed_dim=384,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/ibm-granite/granite-embedding-107m-multilingual",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    adapted_from=None,
+    superseded_by=None,
+)
+
+granite_278m_multilingual = ModelMeta(
+    loader=partial(  # type: ignore
+        sentence_transformers_loader,
+        model_name="ibm-granite/granite-embedding-278m-multilingual",
+        revision="84e3546b88b0cb69f8078608a1df558020bcbf1f",
+    ),
+    name="ibm-granite/granite-embedding-278m-multilingual",
+    languages=GRANITE_LANGUAGES,
+    open_weights=True,
+    revision="84e3546b88b0cb69f8078608a1df558020bcbf1f",
+    release_date="2024-12-18",
+    n_parameters=278_000_000,
+    memory_usage=None,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/ibm-granite/granite-embedding-278m-multilingual",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    adapted_from=None,
+    superseded_by=None,
+)
+
+granite_30m_english = ModelMeta(
+    loader=partial(  # type: ignore
+        sentence_transformers_loader,
+        model_name="ibm-granite/granite-embedding-30m-english",
+        revision="eddbb57470f896b5f8e2bfcb823d8f0e2d2024a5",
+    ),
+    name="ibm-granite/granite-embedding-30m-english",
+    languages=["eng_Latn"],
+    open_weights=True,
+    revision="eddbb57470f896b5f8e2bfcb823d8f0e2d2024a5",
+    release_date="2024-12-18",
+    n_parameters=30_000_000,
+    memory_usage=None,
+    embed_dim=384,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/ibm-granite/granite-embedding-30m-english",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    adapted_from=None,
+    superseded_by=None,
+)
+
+granite_125m_english = ModelMeta(
+    loader=partial(  # type: ignore
+        sentence_transformers_loader,
+        model_name="ibm-granite/granite-embedding-125m-english",
+        revision="e48d3a5b47eaa18e3fe07d4676e187fd80f32730",
+    ),
+    name="ibm-granite/granite-embedding-125m-english",
+    languages=["eng_Latn"],
+    open_weights=True,
+    revision="e48d3a5b47eaa18e3fe07d4676e187fd80f32730",
+    release_date="2024-12-18",
+    n_parameters=125_000_000,
+    memory_usage=None,
+    embed_dim=768,
+    license="apache-2.0",
+    max_tokens=512,
+    reference="https://huggingface.co/ibm-granite/granite-embedding-125m-english",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    adapted_from=None,
+    superseded_by=None,
+)

mteb/models/overview.py

@@ -21,6 +21,7 @@
     google_models,
     gritlm_models,
     gte_models,
+    ibm_granite_models,
     jina_models,
     linq_models,
     llm2vec_models,
@@ -56,6 +57,7 @@
     google_models,
     gritlm_models,
     gte_models,
+    ibm_granite_models,
     jina_models,
     linq_models,
     llm2vec_models,

mteb/models/rerankers_monot5_based.py

@@ -94,8 +94,10 @@ def get_prediction_tokens(
                 token_true_id = tokenizer.get_vocab()[token_true]
                 return token_false_id, token_true_id
             else:
-                raise Exception(f"We don't know the indexes for the non-relevant/relevant tokens for\
-                        the checkpoint {model_name_or_path} and you did not provide any.")
+                raise Exception(
+                    f"We don't know the indexes for the non-relevant/relevant tokens for\
+                        the checkpoint {model_name_or_path} and you did not provide any."
+                )
         else:
             token_false_id = tokenizer.get_vocab()[token_false]
             token_true_id = tokenizer.get_vocab()[token_true]