Model comparison

KaLM-embedding-multilingual-mini-instruct-v2.5 vs bge-m3

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: KaLM-embedding-multilingual-mini-instruct-v2.5's KV cache at 32K is 8.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

KaLM-embedding-multilingual-mini-instruct-v2.5bge-m3
Parameters494M567M
Architectureqwen2bert
Layers2424
Native context131,0728,194
Mixture of expertsnono
Quantizations published11
Smallest quantization0.49 GiB0.59 GiB
Q4_K_M
Licenceapache-2.0mit

KV cache by context

the term that decides long-context viability
ContextKaLM-embedding-multilingual-mini-instruct-v2.5bge-m3Ratio
4,0960.05 GiB0.38 GiB8.00×
8,1920.09 GiB0.75 GiB8.00×
16,3840.19 GiB1.50 GiB8.00×
32,7680.38 GiB3.00 GiB8.00×
65,5360.75 GiB6.00 GiB8.00×
131,0721.50 GiB12.00 GiB8.00×