H

KaLM Embedding Multilingual Mini Instruct v1.5

Dec 2024
HKaLM Embedding Multilingual Mini Instruct v1.5
Compare in
on
self-host onlyprice
51.6BEIR
Dec 2024released
896Dimensions
33K tokensMax input

Rankingconfiguration unstated?

Full ranking →
1Giga Embeddings Instruct 10B A1.8B 082660.32Giga Embeddings Instruct 3B 082657.23Giga Embeddings Instruct57.14BERTA56.55Giga Embeddings Instruct 480M 082654.8
···
40USER Base37.741KaLM Embedding Multilingual Mini Instruct v1.537.742Solon Embeddings Large 0.137.4

Alternatives?

ModelScoreCostPriceWhat changes
BetterEmbeddingGemma 300MunstatedOpen+8%—$0.002Max input: 33K tokens → 2K tokensaOpenGiga Embeddings Instruct 10B A1.8B 0826unstated+60%——Max input: 33K tokens → 8K tokens+1BetterQwen 3 Embedding 8BunstatedOpen+18%—$0.01Memory (4-bit): 0.3 GB → 4.5 GBiBetterMultilingual E5 Large InstructunstatedOpen+9%—$0.01Max input: 33K tokens → 1K tokensBetterQwen 3 Embedding 0.6BunstatedOpen+6%—$0.01Memory (4-bit): 0.3 GB → 0.4 GB

Run it yourself

estimate
Fits on✓ 8 GB (phone, laptop) · full quality✓ 16 GB · full quality✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceMITfree for commercial use

History

3 versions
36.53737.5Release dateMTEB Russian · Multilabel Classificationv1v1.5
36.53737.5Release dateMTEB Russian · Multilabel Classificationv1v1.5

Benchmarks

in Embeddings & Retrieval
BEIR · Mean over tasks · English51.6#86
BEIR · Mean over task types · English51.6#86
BEIR · Retrieval · English51.6#86
MTEB Mandarin Chinese v1 · STS · Mandarin Chinese48.4#50
MTEB Mandarin Chinese v1 · Clustering · Mandarin Chinese57.3#39
MTEB Mandarin Chinese v1 · Reranking · Mandarin Chinese64.4#47
MTEB Mandarin Chinese v1 · Retrieval · Mandarin Chinese70.1#48
MTEB Mandarin Chinese v1 · Pair Classification · Mandarin Chinese72.9#40
MTEB French v1 · Retrieval · multilingual (2 languages)54.3#30
MTEB French v1 · Reranking · French80.6#32
MTEB French v1 · Summarization · French29.3#117
MTEB Polish v1 · Clustering · Polish44.3#20
MTEB Polish v1 · STS · Polish67#33
MTEB Polish v1 · Pair Classification · Polish79.1#63
MTEB Medical v1 · Mean over tasks · multilingual (10 languages)59.7#23
MTEB Medical v1 · Mean over task types · multilingual (10 languages)61.9#17
MTEB Medical v1 · Retrieval · multilingual (10 languages)61.2#26
MTEB Medical v1 · Clustering · English41.4#47
MTEB Medical v1 · Reranking · Mandarin Chinese83.1#45
MTEB Spanish v1 · Classification · Spanish88.5#22
MTEB Russian v1.1 · Clustering · Russian55.9#42
MTEB Russian v1.1 · Reranking · Russian63.7#37
MTEB Russian v1.1 · Multilabel Classification · Russian37.7#41
MTEB Russian v1.1 · STS · Russian70.7#48
MTEB Russian v1.1 · Retrieval · Russian63.9#31
MTEB Russian v1.1 · Pair Classification · Russian55.2#89
RTEB French beta · Mean over tasks · multilingual (2 languages)36.1#80

FAQ

How much does KaLM Embedding Multilingual Mini Instruct v1.5 cost?›

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

Is KaLM Embedding Multilingual Mini Instruct v1.5 free to use?›

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

How good is KaLM Embedding Multilingual Mini Instruct v1.5, and what is it best at?›

#84 of 196 in Embeddings.

Is KaLM Embedding Multilingual Mini Instruct v1.5 better than USER Base?›

On MTEB Russian v1.1 · Multilabel Classification · Russian, KaLM Embedding Multilingual Mini Instruct v1.5 scores 37.7 vs 37.7 — about the same.

When was KaLM Embedding Multilingual Mini Instruct v1.5 released? Is there a newer version?›

Released 2024-12-26. It is the current version in its line.

Is KaLM Embedding Multilingual Mini Instruct v1.5 open source?›

Yes, the weights are open (MIT licence).

Can I run KaLM Embedding Multilingual Mini Instruct v1.5 locally?›

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

What is KaLM Embedding Multilingual Mini Instruct v1.5?›

KaLM Embedding Multilingual Mini Instruct v1.5 is an open-weights embeddings & retrieval model from HIT-TMG, released Dec 2024.

What are the best alternatives to KaLM Embedding Multilingual Mini Instruct v1.5?›

EmbeddingGemma 300M (better), Giga Embeddings Instruct 10B A1.8B 0826 (open weights).