wrong multiplicity for large multiplicity element when using projection based embedding

When I use projection based embedding for a cluster with one Eu(2+) and one water molecule, the total multiplicity is set to 8, and the embedding region (Eu2+) is also 8, like this,

$molecule
 2 8
 --
 2 8
 Eu 3.00000  0.00000  0.00000
 --
 0 1
 O  1.26276 -0.06380  0.00000
 H  1.70602  0.31170 -0.76717 
 H  1.70602  0.31170  0.76717 
$end  

$rem 
    jobtype sp 
    method pbe0
    env_method pbe 
    basis def2-svp
    ecp def2-ecp 
    unrestricted true
    gen_scfman_embed true
$end

however, the embedding code seems can not deal with such a large multiplicity, it gives,

Embedding calculation for unrestricted system
NBasis for each fragment
        78
        24
NOcc for each fragment (alpha)
        17
         5
NOcc for each fragment (beta)
        16
         5
Number of orbitals that are fixed: 5 (alpha), 5 (beta)
Read converged orbitals from disk (alpha)
Localizing occupieds...
 Performing Pipek-Mezey Localization.
 Localization complete after     6 iterations.
Occ 1 is assigned to frgm 2
Occ 2 is assigned to frgm 1
Occ 3 is assigned to frgm 1
Occ 4 is assigned to frgm 1
Occ 5 is assigned to frgm 1
Occ 6 is assigned to frgm 1
Occ 7 is assigned to frgm 1
Occ 8 is assigned to frgm 1
Occ 9 is assigned to frgm 1
Occ 10 is assigned to frgm 1
Occ 11 is assigned to frgm 1
Occ 12 is assigned to frgm 2
Occ 13 is assigned to frgm 1
Occ 14 is assigned to frgm 1
Occ 15 is assigned to frgm 1
Occ 16 is assigned to frgm 2
Occ 17 is assigned to frgm 2
Occ 18 is assigned to frgm 2
Occ 19 is assigned to frgm 0
Occ 20 is assigned to frgm 0
Occ 21 is assigned to frgm 0
Occ 22 is assigned to frgm 1
Occ 23 is assigned to frgm 1
Occ 24 is assigned to frgm 1
Occ 25 is assigned to frgm 1
remove work dirs /tmp/qchem2910.0 -- /tmp/qchem2910.0
rm -rf /tmp/qchem2910.0
rm -rf /tmp/qchem2910

which shows that it split the total 7 alpha electrons into 4 alpha and 3 beta. So how can I use the embedding method correctly?