T

VLM2Vec LoRA

Oct 2024
TVLM2Vec LoRA
Compare in
on
self-host onlyprice
42.7MIEB lite
Oct 2024released
3,072Dimensions
131K tokensMax input
enLanguages

Ranking?

Full ranking →
1LCO Embedding Omni 7B68.82LCO Embedding Omni 3B66.43E5 Omni 7B65.24DNA VL STEER 2B64.35EmbeddingGemma 263.5
···
29CLIP ViT B 16 DataComp.XL S13B B90K46.630VLM2Vec LoRA4631VLM2Vec Full45.9

Strong at

C
Embeddings & Retrieval#59 of 89

Alternatives?

ModelScoreCostPriceWhat changes
BetterEmbeddingGemma 2unstatedOpen+38%—$0.02Max input: 131K tokens → 8K tokens+1LOpenLCO Embedding Omni 7Bunstated+50%——Max input: 131K tokens → 33K tokens+2BetterJina Embeddings v5 Omni NanounstatedOpen+7%—$0.02Max input: 131K tokens → 8K tokens+1BetterJina Embeddings v5 Omni SmallunstatedOpen+33%—$0.05Max input: 131K tokens → 33K tokens+1VBetterVoyage Multimodal 3+30%—$0.12Max input: 131K tokens → 33K tokens+1

Run it yourself

estimate
Fits on✓ 8 GB (phone, laptop) · 8-bit✓ 16 GB · full quality✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use

Benchmarks

in Embeddings & Retrieval
MIEB English · Mean over tasks · multilingual (13 languages)53.4#22
MIEB English · Mean over task types · multilingual (13 languages)53.4#22
MIEB English · Document Understanding · English49.7#52
MIEB English · Zero-shot Classification · English46.3#34
MIEB English · Any-to-any Retrieval · English27.7#33
MIEB English · Visual semantic similarity (English) · multilingual (13 languages)72.6#7
MIEB English · Image Classification · English62#42
MIEB English · Compositionality · English34.6#33
MIEB English · Vision Centric QA · English62#3
MIEB English · Image Clustering · English72.6#33
MIEB Img · Mean over tasks · multilingual (13 languages)56.1#21
MIEB Img · Mean over task types · multilingual (13 languages)56.1#21
MIEB Img · Image Classification · English62#42
MIEB Img · Any-to-any Retrieval · English27.8#38
MIEB Img · Visual semantic similarity (English) · English70.2#8
MIEB Img · Visual semantic similarity (multilingual) · multilingual (13 languages)48#10
MIEB Img · Image Clustering · English72.6#33
MIEB lite · Mean over tasks · multilingual (41 languages)46#30
MIEB lite · Mean over task types · multilingual (41 languages)46#30
MIEB lite · Vision Centric QA · English65.3#5
MIEB lite · Any-to-any Retrieval · English21#42
MIEB lite · Document Understanding · English42.7#59
MIEB lite · Zero-shot Classification · English32.1#44
MIEB lite · Compositionality · English29.4#41
MIEB lite · Image Classification · English64.8#41
MIEB lite · Any-to-any Multilingual Retrieval · multilingual (40 languages)24.8#25
MIEB lite · Image Clustering · English66.3#36
MIEB lite · Visual semantic similarity (English) · English70.9#19
MIEB lite · Visual semantic similarity (multilingual) · multilingual (13 languages)42.2#15
MIEB multilingual · Mean over tasks · multilingual (41 languages)50.5#20
MIEB multilingual · Mean over task types · multilingual (41 languages)50.5#20
MIEB multilingual · Zero-shot Classification · English46.3#34
MIEB multilingual · Image Classification · English62#42
MIEB multilingual · Document Understanding · English49.7#52
MIEB multilingual · Any-to-any Retrieval · English27.7#33
MIEB multilingual · Compositionality · English34.6#33
MIEB multilingual · Vision Centric QA · English62#3
MIEB multilingual · Visual semantic similarity (English) · multilingual (13 languages)72.6#7
MIEB multilingual · Visual semantic similarity (multilingual) · multilingual (13 languages)42.2#15
MIEB multilingual · Any-to-any Multilingual Retrieval · multilingual (40 languages)34.9#15
MIEB multilingual · Image Clustering · English72.6#33

FAQ

How much does VLM2Vec LoRA cost?›

There is no per-unit API price: self-host only.

Is VLM2Vec LoRA free to use?›

There is no free API. The weights are open, so you can run it yourself.

How good is VLM2Vec LoRA, and what is it best at?›

#59 of 89 in Multimodal.

Is VLM2Vec LoRA better than CLIP ViT B 16 DataComp.XL S13B B90K?›

On MIEB lite · Mean over task types · multilingual (41 languages), VLM2Vec LoRA scores 46 vs 46.6 — about the same.

When was VLM2Vec LoRA released? Is there a newer version?›

Released 2024-10-08. It is the current version in its line.

Is VLM2Vec LoRA open source?›

Yes, the weights are open (Apache-2.0 licence).

Can I run VLM2Vec LoRA locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

What is VLM2Vec LoRA?›

VLM2Vec LoRA is an open-weights embeddings & retrieval model from TIGER-Lab, released Oct 2024.

What are the best alternatives to VLM2Vec LoRA?›

EmbeddingGemma 2 (better), LCO Embedding Omni 7B (open weights).