Add model specific dependencies in pyproject.toml

docs/adding_a_model.md

@@ -133,3 +133,17 @@ model = ModelMeta(
    ...
 )
 ```
+
+##### Adding model dependencies in pyproject.toml
+If your are adding a model that requires additional dependencies, you can add them to the `pyproject.toml` file and instead of checking whether dependencies are installed or not make use of `requires_package` from [requires_package.py](../mteb/requires_packages.py). For example:
+
+In the [voyage_models.py](../mteb/models/voyage_models.py) file, we have added the following code:
+```python
+requires_package(self, "voyageai", model_name, "pip install 'mteb[voyageai]'")
+```
+and also updated [pyproject.toml]((../pyproject.toml)) file with the following code:
+```python
+voyageai = ["voyageai>=1.0.0,<2.0.0"]
+```
+so that it will check whether voyageai is installed or not. If not, then it will give an error message to install voyageai. This has done so as to give clear installation warnings. 
+If you want to give suggestion instead of warning, you can use `suggest_package` from [requires_package.py](../mteb/requires_packages.py).

mteb/models/blip2_models.py

@@ -11,17 +11,17 @@
 
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
+from mteb.requires_package import requires_package
 
 
 def blip2_loader(**kwargs):
-    try:  # a temporal fix for the dependency issues.
-        from lavis.models.blip2_models.blip2_image_text_matching import (
-            Blip2ITM,
-        )
-    except ImportError:
-        raise ImportError(
-            "Please install `pip install mteb[blip2]` to use BLIP-2 models."
-        )
+    model_name = kwargs.get("model_name", "BLIP-2")
+    requires_package(
+        blip2_loader, "salesforce-lavis", model_name, "pip install 'mteb[blip2]'"
+    )
+    from lavis.models.blip2_models.blip2_image_text_matching import (
+        Blip2ITM,
+    )
 
     class BLIP2ModelWrapper:
         def __init__(

mteb/models/bm25.py

@@ -6,18 +6,16 @@
 from mteb.evaluation.evaluators.RetrievalEvaluator import DRESModel
 from mteb.model_meta import ModelMeta
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package
 
 logger = logging.getLogger(__name__)
 
 
 def bm25_loader(**kwargs):
-    try:
-        import bm25s
-        import Stemmer
-    except ImportError:
-        raise ImportError(
-            "bm25s or PyStemmer is not installed. Please install it with `pip install mteb[bm25s]`."
-        )
+    model_name = kwargs.get("model_name", "BM25")
+    requires_package(bm25_loader, "bm25s", model_name, "pip install mteb[bm25s]")
+    import bm25s
+    import Stemmer
 
     class BM25Search(DRESModel, Wrapper):
         """BM25 search"""

mteb/models/cohere_v.py

@@ -14,14 +14,15 @@
 
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
-from mteb.requires_package import requires_image_dependencies
+from mteb.requires_package import requires_image_dependencies, requires_package
 
 
 def cohere_v_loader(**kwargs):
-    try:
-        import cohere
-    except ImportError:
-        raise ImportError("To use cohere models, please run `pip install cohere`.")
+    model_name = kwargs.get("model_name", "Cohere")
+    requires_package(
+        cohere_v_loader, "cohere", model_name, "pip install 'mteb[cohere]'"
+    )
+    import cohere
 
     class CohereMultiModalModelWrapper:
         def __init__(

mteb/models/colbert_models.py

@@ -11,6 +11,7 @@
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package
 
 logger = logging.getLogger(__name__)
 
@@ -34,12 +35,8 @@ def __init__(
                 and finally to the specific prompt type.
             **kwargs: Additional arguments to pass to the model.
         """
-        try:
-            from pylate import models as colbert_model
-        except ModuleNotFoundError as e:
-            raise ModuleNotFoundError(
-                "To use the ColBERT models `pylate` is required. Please install it with `pip install mteb[pylate]`."
-            ) from e
+        requires_package(self, "pylate", model_name, "pip install mteb[pylate]")
+        from pylate import models as colbert_model
 
         self.model_name = model_name
         self.model = colbert_model.ColBERT(self.model_name, revision=revision, **kwargs)

mteb/models/google_models.py

@@ -9,6 +9,7 @@
 from mteb.encoder_interface import Encoder, PromptType
 from mteb.model_meta import ModelMeta
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package
 
 MULTILINGUAL_EVALUATED_LANGUAGES = [
     "arb_Arab",
@@ -74,12 +75,11 @@ def _embed(
         """Embeds texts with a pre-trained, foundational model.
         From https://cloud.google.com/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings#generative-ai-get-text-embedding-python_vertex_ai_sdk
         """
-        try:
-            from vertexai.language_models import TextEmbeddingInput, TextEmbeddingModel
-        except ImportError:
-            raise ImportError(
-                "The `vertexai` package is required to run the google API, please install it using `pip install vertexai`"
-            )
+        requires_package(
+            self, "vertexai", self.model_name, "pip install 'mteb[vertexai]'"
+        )
+        from vertexai.language_models import TextEmbeddingInput, TextEmbeddingModel
+
         model = TextEmbeddingModel.from_pretrained(self.model_name)
         if titles:
             # Allow title-only embeddings by replacing text with a space

mteb/models/instruct_wrapper.py

@@ -10,6 +10,7 @@
 
 from mteb.encoder_interface import PromptType
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package
 
 logger = logging.getLogger(__name__)
 
@@ -20,12 +21,10 @@ def instruct_wrapper(
     instruction_template: str | Callable[[str], str] | None = None,
     **kwargs,
 ):
-    try:
-        from gritlm import GritLM
-    except ImportError:
-        raise ImportError(
-            f"Please install `pip install mteb[gritlm]` to use {model_name_or_path}."
-        )
+    requires_package(
+        instruct_wrapper, "gritlm", model_name_or_path, "pip install 'mteb[gritlm]'"
+    )
+    from gritlm import GritLM
 
     class InstructWrapper(GritLM, Wrapper):
         def __init__(

mteb/models/jina_models.py

@@ -12,6 +12,7 @@
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
 from mteb.models.sentence_transformer_wrapper import SentenceTransformerWrapper
+from mteb.requires_package import requires_package
 
 logger = logging.getLogger(__name__)
 
@@ -140,19 +141,14 @@ def __init__(
             raise RuntimeError(
                 f"sentence_transformers version {st_version} is lower than the required version 3.1.0"
             )
-        try:
-            import einops  # noqa: F401
-        except ImportError:
-            raise ImportError(
-                "To use the jina-embeddings-v3 models `einops` is required. Please install it with `pip install mteb[jina]`."
-            )
-        try:
-            import flash_attn  # noqa: F401
-        except ImportError:
-            logger.warning(
-                "Using flash_attn for jina-embeddings-v3 models is recommended. Please install it with `pip install mteb[flash_attention]`."
-                "Fallback to native implementation."
-            )
+        requires_package(self, "jina", model, "pip install 'mteb[jina]'")
+        import einops  # noqa: F401
+
+        requires_package(
+            self, "flash_attention", model, "pip install 'mteb[flash_attention]'"
+        )
+        import flash_attn  # noqa: F401
+
         super().__init__(model, revision, model_prompts, **kwargs)
 
     def encode(

mteb/models/llm2clip_models.py

@@ -12,7 +12,7 @@
 
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
-from mteb.requires_package import requires_image_dependencies
+from mteb.requires_package import requires_image_dependencies, requires_package
 
 MODEL2PROCESSOR = {
     "microsoft/LLM2CLIP-Openai-L-14-336": "openai/clip-vit-large-patch14-336",
@@ -22,13 +22,11 @@
 
 
 def llm2clip_loader(**kwargs):
-    try:
-        from llm2vec import LLM2Vec
-    except ImportError:
-        # https://github.com/baaivision/EVA/tree/master/EVA-CLIP#setup
-        raise ImportError(
-            "To use the LLM2CLIP models `llm2vec` is required. Please install it with `pip install llm2vec`."
-        )
+    model_name = kwargs.get("model_name", "LLM2CLIP")
+    requires_package(
+        llm2clip_loader, "llm2vec", model_name, "pip install 'mteb[llm2vec]'"
+    )
+    from llm2vec import LLM2Vec
 
     class LLM2CLIPWrapper:
         def __init__(

mteb/models/llm2vec_models.py

@@ -9,6 +9,7 @@
 from mteb.encoder_interface import Encoder, PromptType
 from mteb.model_meta import ModelMeta
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package, suggest_package
 
 logger = logging.getLogger(__name__)
 
@@ -56,21 +57,21 @@ def __init__(
         *args,
         **kwargs,
     ):
-        try:
-            from llm2vec import LLM2Vec
-        except ImportError:
-            raise ImportError(
-                "To use the LLM2Vec models `llm2vec` is required. Please install it with `pip install llm2vec`."
-            )
+        model_name = kwargs.get("model_name", "LLM2Vec")
+        requires_package(self, "llm2vec", model_name, "pip install 'mteb[llm2vec]'")
+        from llm2vec import LLM2Vec
+
         extra_kwargs = {}
-        try:
-            import flash_attn  # noqa
+        if suggest_package(
+            self,
+            "flash_attn",
+            model_name,
+            "pip install flash-attn --no-build-isolation",
+        ):
+            import flash_attn  # noqa: F401
 
             extra_kwargs["attn_implementation"] = "flash_attention_2"
-        except ImportError:
-            logger.warning(
-                "LLM2Vec models were trained with flash attention enabled. For optimal performance, please install the `flash_attn` package with `pip install flash-attn --no-build-isolation`."
-            )
+
         self.model_prompts = (
             self.validate_task_to_prompt_name(model_prompts) if model_prompts else None
         )

mteb/models/moco_models.py

@@ -10,14 +10,13 @@
 
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
-from mteb.requires_package import requires_image_dependencies
+from mteb.requires_package import requires_image_dependencies, requires_package
 
 
 def mocov3_loader(**kwargs):
-    try:
-        import timm
-    except ImportError:
-        raise ImportError("Please install `pip install timm` to use MOCOv3 models.")
+    model_name = kwargs.get("model_name", "MOCOv3")
+    requires_package(mocov3_loader, "timm", model_name, "pip install 'mteb[timm]'")
+    import timm
 
     class MOCOv3Wrapper:
         """A wrapper class for MOCOv3 models that supports image encoding.

mteb/models/model2vec_models.py

@@ -10,6 +10,7 @@
 from mteb.model_meta import ModelMeta
 from mteb.models.bge_models import bge_training_data
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package
 
 logger = logging.getLogger(__name__)
 
@@ -26,12 +27,8 @@ def __init__(
             model_name: The Model2Vec model to load from HuggingFace Hub.
             **kwargs: Additional arguments to pass to the wrapper.
         """
-        try:
-            from model2vec import StaticModel  # type: ignore
-        except ModuleNotFoundError as e:
-            raise ModuleNotFoundError(
-                "To use the Model2Vec models `model2vec` is required. Please install it with `pip install mteb[model2vec]`."
-            ) from e
+        requires_package(self, "model2vec", model_name, "pip install 'mteb[model2vec]'")
+        from model2vec import StaticModel  # type: ignore
 
         self.model_name = model_name
         self.static_model = StaticModel.from_pretrained(self.model_name)

mteb/models/openai_models.py

@@ -29,16 +29,16 @@ def __init__(
         requires_package(
             self,
             "openai",
-            "Openai text embedding",
-            install_instruction="pip install mteb[openai]",
+            model_name,
+            install_instruction="pip install 'mteb[openai]'",
         )
         from openai import OpenAI
 
         requires_package(
             self,
             "tiktoken",
-            "Tiktoken package",
-            install_instruction="pip install mteb[openai]",
+            model_name,
+            install_instruction="pip install 'mteb[openai]'",
         )
         import tiktoken
 

mteb/models/openclip_models.py

@@ -10,14 +10,18 @@
 
 from mteb.encoder_interface import PromptType
 from mteb.model_meta import ModelMeta
-from mteb.requires_package import requires_image_dependencies
+from mteb.requires_package import requires_image_dependencies, requires_package
 
 
 def openclip_loader(**kwargs):
-    try:
-        import open_clip
-    except ImportError:
-        raise ImportError("Please run `pip install open_clip_torch`.")
+    model_name = kwargs.get("model_name", "CLIP-ViT")
+    requires_package(
+        openclip_loader,
+        "open_clip_torch",
+        model_name,
+        "pip install 'mteb[open_clip_torch]'",
+    )
+    import open_clip
 
     class OpenCLIPWrapper:
         def __init__(

mteb/models/repllama_models.py

@@ -12,6 +12,7 @@
 from mteb.encoder_interface import Encoder, PromptType
 from mteb.model_meta import ModelMeta
 from mteb.models.wrapper import Wrapper
+from mteb.requires_package import requires_package
 
 logger = logging.getLogger(__name__)
 
@@ -26,12 +27,10 @@ def __init__(
         model_prompts: dict[str, str] | None = None,
         **kwargs,
     ):
-        try:
-            from peft import PeftModel
-        except ImportError:
-            raise ImportError(
-                "To use the RepLLaMA based models `peft` is required. Please install it with `pip install 'mteb[peft]'`."
-            )
+        requires_package(
+            self, "peft", peft_model_name_or_path, "pip install 'mteb[peft]'"
+        )
+        from peft import PeftModel
 
         self.base_model = AutoModel.from_pretrained(
             base_model_name_or_path,