fix: model memory

mteb/models/model_implementations/google_models.py

@@ -275,5 +275,5 @@ def gemma_embedding_loader(model_name: str, revision: str, **kwargs):
     public_training_data=None,
     training_datasets=GECKO_TRAINING_DATA,
     similarity_fn_name="cosine",
-    memory_usage_mb=578,
+    memory_usage_mb=1155,
 )

mteb/models/model_implementations/nb_sbert.py

@@ -11,7 +11,7 @@
     revision="b95656350a076aeafd2d23763660f80655408cc6",
     release_date="2022-11-23",
     n_parameters=1_780_000_000,
-    memory_usage_mb=197,
+    memory_usage_mb=678,
     embed_dim=4096,
     license="apache-2.0",
     max_tokens=75,

mteb/models/model_implementations/nvidia_llama_nemoretriever_colemb.py

@@ -146,7 +146,7 @@ def encode(
     release_date="2025-06-27",
     modalities=["image", "text"],
     n_parameters=2_418_000_000,
-    memory_usage_mb=9224,
+    memory_usage_mb=4610,
     max_tokens=8192,
     embed_dim=2048,
     license="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1/blob/main/LICENSE",
@@ -172,7 +172,7 @@ def encode(
     release_date="2025-06-27",
     modalities=["image", "text"],
     n_parameters=4_407_000_000,
-    memory_usage_mb=16811,
+    memory_usage_mb=8403,
     max_tokens=8192,
     embed_dim=3072,
     license="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1/blob/main/LICENSE",

mteb/models/model_implementations/nvidia_models.py

@@ -146,7 +146,7 @@ def instruction_template(
     revision="570834afd5fef5bf3a3c2311a2b6e0a66f6f4f2c",
     release_date="2024-09-13",  # initial commit of hf model.
     n_parameters=7_850_000_000,
-    memory_usage_mb=29945,
+    memory_usage_mb=14975,
     embed_dim=4096,
     license="cc-by-nc-4.0",
     max_tokens=32768,

mteb/models/model_implementations/ops_moa_models.py

@@ -27,7 +27,7 @@ def encode(self, sentences: list[str], **kwargs) -> np.ndarray:
     languages=["zho-Hans"],
     loader=OPSWrapper,
     n_parameters=int(343 * 1e6),
-    memory_usage_mb=2e3,
+    memory_usage_mb=1308,
     max_tokens=512,
     embed_dim=1536,
     license="cc-by-nc-4.0",
@@ -58,7 +58,7 @@ def encode(self, sentences: list[str], **kwargs) -> np.ndarray:
     languages=["zho-Hans"],
     loader=OPSWrapper,
     n_parameters=int(343 * 1e6),
-    memory_usage_mb=2e3,
+    memory_usage_mb=1242,
     max_tokens=512,
     embed_dim=1536,
     license="cc-by-nc-4.0",

mteb/models/model_implementations/promptriever_models.py

@@ -80,7 +80,7 @@ def loader_inner(**kwargs: Any) -> EncoderProtocol:
     revision="01c7f73d771dfac7d292323805ebc428287df4f9-30b14e3813c0fa45facfd01a594580c3fe5ecf23",  # base-peft revision
     release_date="2024-09-15",
     n_parameters=7_000_000_000,
-    memory_usage_mb=27,
+    memory_usage_mb=26703,
     max_tokens=4096,
     embed_dim=4096,
     license="apache-2.0",
@@ -115,7 +115,7 @@ def loader_inner(**kwargs: Any) -> EncoderProtocol:
     },
     release_date="2024-09-15",
     n_parameters=8_000_000_000,
-    memory_usage_mb=31,
+    memory_usage_mb=30518,
     max_tokens=8192,
     embed_dim=4096,
     license="apache-2.0",
@@ -143,7 +143,7 @@ def loader_inner(**kwargs: Any) -> EncoderProtocol:
     revision="5206a32e0bd3067aef1ce90f5528ade7d866253f-8b677258615625122c2eb7329292b8c402612c21",  # base-peft revision
     release_date="2024-09-15",
     n_parameters=8_000_000_000,
-    memory_usage_mb=31,
+    memory_usage_mb=30518,
     max_tokens=8192,
     embed_dim=4096,
     training_datasets={
@@ -175,7 +175,7 @@ def loader_inner(**kwargs: Any) -> EncoderProtocol:
     revision="7231864981174d9bee8c7687c24c8344414eae6b-876d63e49b6115ecb6839893a56298fadee7e8f5",  # base-peft revision
     release_date="2024-09-15",
     n_parameters=7_000_000_000,
-    memory_usage_mb=27,
+    memory_usage_mb=26703,
     training_datasets={
         # "samaya-ai/msmarco-w-instructions",
         "mMARCO-NL",  # translation not trained on

mteb/models/model_implementations/qwen3_models.py

@@ -139,7 +139,7 @@ def q3e_instruct_loader(
     revision="b22da495047858cce924d27d76261e96be6febc0",  # Commit of @tomaarsen
     release_date="2025-06-05",
     n_parameters=595776512,
-    memory_usage_mb=2272,
+    memory_usage_mb=1136,
     embed_dim=1024,
     max_tokens=32768,
     license="apache-2.0",
@@ -161,7 +161,7 @@ def q3e_instruct_loader(
     revision="636cd9bf47d976946cdbb2b0c3ca0cb2f8eea5ff",  # Commit of @tomaarsen
     release_date="2025-06-05",
     n_parameters=4021774336,
-    memory_usage_mb=15341,
+    memory_usage_mb=7671,
     embed_dim=2560,
     max_tokens=32768,
     license="apache-2.0",
@@ -183,7 +183,7 @@ def q3e_instruct_loader(
     revision="4e423935c619ae4df87b646a3ce949610c66241c",  # Commit of @tomaarsen
     release_date="2025-06-05",
     n_parameters=7567295488,
-    memory_usage_mb=28866,
+    memory_usage_mb=14433,
     embed_dim=4096,
     max_tokens=32768,
     license="apache-2.0",

mteb/models/model_implementations/qzhou_models.py

@@ -63,7 +63,7 @@ def instruction_template(
     revision="f1e6c03ee3882e7b9fa5cec91217715272e433b8",
     release_date="2025-08-24",
     n_parameters=7_070_619_136,
-    memory_usage_mb=29070,
+    memory_usage_mb=14436,
     embed_dim=3584,
     license="apache-2.0",
     max_tokens=8192,