model: Add ReasonIR

mteb/models/overview.py

@@ -84,6 +84,7 @@
     qtack_models,
     qwen3_models,
     qzhou_models,
+    reasonir_model,
     repllama_models,
     rerankers_custom,
     rerankers_monot5_based,
@@ -178,6 +179,7 @@
     qtack_models,
     qwen3_models,
     qzhou_models,
+    reasonir_model,
     repllama_models,
     rerankers_custom,
     rerankers_monot5_based,

mteb/models/reasonir_model.py

@@ -0,0 +1,59 @@
+from __future__ import annotations
+
+from functools import partial
+
+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 (
+        # https://github.com/facebookresearch/ReasonIR/blob/0aac96269e455965949df16520fab72da68ffc22/evaluation/bright/configs/reasonir/economics.json#L3
+        f"<|user|>\n{instruction}<|embed|>\n"
+        if (prompt_type is None or prompt_type == PromptType.query) and instruction
+        else "<|embed|>\n"
+    )
+
+
+REASONIR_TRAINING_DATA = {
+    # source, section D: https://arxiv.org/pdf/2504.20595
+    "MSMARCO": ["train"],
+    "NQ": ["train"],
+    "FEVER": ["train"],
+    "HotpotQA": ["train"],
+    "MIRACLRetrieval": ["train"],
+    "MrTidyRetrieval": ["train"],
+    "T2Reranking": ["train"],
+    "DuRetrieval": ["train"],
+    "QuoraRetrieval": ["train"],
+}
+
+ReasonIR_8B = ModelMeta(
+    loader=partial(
+        InstructSentenceTransformerWrapper,
+        model_name="ReasonIR/ReasonIR-8B",
+        revision="c3d0690370ff4a8c3d3882d8dfa85c43650034fa",
+        instruction_template=instruction_template,
+        trust_remote_code=True,
+    ),
+    name="ReasonIR/ReasonIR-8B",
+    languages=["eng-Latn"],
+    open_weights=True,
+    revision="c3d0690370ff4a8c3d3882d8dfa85c43650034fa",
+    release_date="2025-04-29",
+    n_parameters=7_500_000_000,
+    memory_usage_mb=None,
+    embed_dim=4096,
+    license="cc-by-nc-4.0",
+    max_tokens=131072,
+    reference="https://huggingface.co/ReasonIR/ReasonIR-8B",
+    similarity_fn_name="cosine",
+    framework=["Sentence Transformers", "PyTorch"],
+    use_instructions=True,
+    training_datasets=REASONIR_TRAINING_DATA,
+    public_training_code="https://github.com/facebookresearch/ReasonIR/tree/main/training",
+    public_training_data="https://huggingface.co/datasets/reasonir/reasonir-data",
+)