Assuming a text-vector file with header line, this will help you select the number of vectors you want, and clean a little.

vector_select.sh

# Number of vector lines you want
N=50000

IN=$1
OUT=$2

# Get the dimension from the header.
DIM=$(head -n 1 $IN | cut -d' ' -f2)

# Actually take the top...
head -n $(($N+1)) $IN > $OUT

# Fix the header
sed -i.bak "1 s/^.*$/${N} ${DIM}/" $OUT