model: Add BMRetriever

mteb/models/bmretriever_models.py

@@ -0,0 +1,185 @@
+from __future__ import annotations
+
+from functools import partial
+from typing import Any, Callable
+
+import torch
+from sentence_transformers import SentenceTransformer
+from sentence_transformers.models import Pooling, Transformer
+
+from mteb.encoder_interface import PromptType
+from mteb.model_meta import ModelMeta
+from mteb.models.instruct_wrapper import InstructSentenceTransformerWrapper
+
+
+def instruction_template(
+    instruction: str, prompt_type: PromptType | None = None
+) -> str:
+    return (
+        f"{instruction}\nQuery: "
+        if (prompt_type is None or prompt_type == PromptType.query) and instruction
+        else "Represent this passage\npassage: "
+    )
+
+
+class BMRetrieverWrapper(InstructSentenceTransformerWrapper):
+    def __init__(
+        self,
+        model_name: str,
+        instruction_template: Callable[[str, PromptType | None], str] | None = None,
+        max_seq_length: int | None = None,
+        apply_instruction_to_passages: bool = True,
+        padding_side: str | None = None,
+        add_eos_token: bool = False,
+        prompts_dict: dict[str, str] | None = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.instruction_template = instruction_template
+        self.apply_instruction_to_passages = apply_instruction_to_passages
+        self.add_eos_token = add_eos_token
+        self.prompts_dict = prompts_dict
+
+        transformer = Transformer(
+            model_name,
+            max_seq_length=max_seq_length,
+            **kwargs,
+        )
+        pooling = Pooling(
+            transformer.get_word_embedding_dimension(), pooling_mode="lasttoken"
+        )
+        self.model = SentenceTransformer(modules=[transformer, pooling])
+
+        if max_seq_length is not None:
+            self.model.max_seq_length = max_seq_length
+
+        if padding_side is not None:
+            self.model.tokenizer.padding_side = padding_side
+
+
+# https://huggingface.co/datasets/BMRetriever/biomed_retrieval_dataset
+BMRETRIEVER_TRAINING_DATA = {
+    "FEVER": ["train"],
+    "MSMARCO": ["train"],
+    "NQ": ["train"],
+}
+
+BMRetriever_410M = ModelMeta(
+    loader=partial(
+        BMRetrieverWrapper,
+        model_name="BMRetriever/BMRetriever-410M",
+        config_args={"revision": "e3569bfbcfe3a1bc48c142e11a7b0f38e86065a3"},
+        model_args={"torch_dtype": torch.float32},
+        instruction_template=instruction_template,
+        padding_side="left",
+        add_eos_token=True,
+        apply_instruction_to_passages=True,
+    ),
+    name="BMRetriever/BMRetriever-410M",
+    languages=["eng-Latn"],
+    open_weights=True,
+    revision="e3569bfbcfe3a1bc48c142e11a7b0f38e86065a3",
+    release_date="2024-04-29",
+    embed_dim=1024,
+    n_parameters=353_822_720,
+    memory_usage_mb=1349,
+    max_tokens=2048,
+    license="mit",
+    reference="https://huggingface.co/BMRetriever/BMRetriever-410M",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=True,
+    public_training_code=None,
+    public_training_data=None,
+    training_datasets=BMRETRIEVER_TRAINING_DATA,
+)
+
+BMRetriever_1B = ModelMeta(
+    loader=partial(
+        BMRetrieverWrapper,
+        model_name="BMRetriever/BMRetriever-1B",
+        config_args={"revision": "1b758c5f4d3af48ef6035cc4088bdbcd7df43ca6"},
+        model_args={"torch_dtype": torch.float32},
+        instruction_template=instruction_template,
+        padding_side="left",
+        add_eos_token=True,
+        apply_instruction_to_passages=True,
+    ),
+    name="BMRetriever/BMRetriever-1B",
+    languages=["eng-Latn"],
+    open_weights=True,
+    revision="1b758c5f4d3af48ef6035cc4088bdbcd7df43ca6",
+    release_date="2024-04-29",
+    embed_dim=2048,
+    n_parameters=908_759_040,
+    memory_usage_mb=3466,
+    max_tokens=2048,
+    license="mit",
+    reference="https://huggingface.co/BMRetriever/BMRetriever-1B",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=True,
+    public_training_code=None,
+    public_training_data=None,
+    training_datasets=BMRETRIEVER_TRAINING_DATA,
+)
+
+BMRetriever_2B = ModelMeta(
+    loader=partial(
+        BMRetrieverWrapper,
+        model_name="BMRetriever/BMRetriever-2B",
+        config_args={"revision": "718179afd57926369c347f46eee616db81084941"},
+        model_args={"torch_dtype": torch.float32},
+        instruction_template=instruction_template,
+        padding_side="left",
+        add_eos_token=True,
+        apply_instruction_to_passages=True,
+    ),
+    name="BMRetriever/BMRetriever-2B",
+    languages=["eng-Latn"],
+    open_weights=True,
+    revision="718179afd57926369c347f46eee616db81084941",
+    release_date="2024-04-29",
+    embed_dim=2048,
+    n_parameters=2_506_172_416,
+    memory_usage_mb=9560,
+    max_tokens=8192,
+    license="mit",
+    reference="https://huggingface.co/BMRetriever/BMRetriever-2B",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=True,
+    public_training_code=None,
+    public_training_data=None,
+    training_datasets=BMRETRIEVER_TRAINING_DATA,
+)
+
+BMRetriever_7B = ModelMeta(
+    loader=partial(
+        BMRetrieverWrapper,
+        model_name="BMRetriever/BMRetriever-7B",
+        config_args={"revision": "e3569bfbcfe3a1bc48c142e11a7b0f38e86065a3"},
+        model_args={"torch_dtype": torch.float32},
+        instruction_template=instruction_template,
+        padding_side="left",
+        add_eos_token=True,
+        apply_instruction_to_passages=True,
+    ),
+    name="BMRetriever/BMRetriever-7B",
+    languages=["eng-Latn"],
+    open_weights=True,
+    revision="e3569bfbcfe3a1bc48c142e11a7b0f38e86065a3",
+    release_date="2024-04-29",
+    embed_dim=4096,
+    n_parameters=7_110_660_096,
+    memory_usage_mb=27124,
+    max_tokens=32768,
+    license="mit",
+    reference="https://huggingface.co/BMRetriever/BMRetriever-7B",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=True,
+    public_training_code=None,
+    public_training_data=None,
+    training_datasets=BMRETRIEVER_TRAINING_DATA,
+)

mteb/models/overview.py

@@ -21,6 +21,7 @@
     blip2_models,
     blip_models,
     bm25,
+    bmretriever_models,
     cadet_models,
     cde_models,
     clip_models,
@@ -121,6 +122,7 @@
     blip2_models,
     blip_models,
     bm25,
+    bmretriever_models,
     cadet_models,
     clip_models,
     codesage_models,