Let $\mathcal{M}_{n,m}(\mathbb{F_q})$ be the set of $n$ by $m$ matrices over the finite field of order $q$, with $n \geq m$.
For a (non-trivial) subspace $V \subset \mathcal{M}_{n,m}(\mathbb{F_q})$, we denote by $f(V) := \frac{|\{M \in V \, \setminus \, \{0\} \, : \, M \text{ does not have full rank} \}|}{|V \, \setminus \, {0}|}$ the fraction of non-zero matrices in $V$ which does not have full rank.
For $V$ chosen uniformly at random of dimension $d$, the average of $f(V)$ is equal to $f(\mathcal{M}_{n,m}(\mathbb{F_q})) \approx q^{m-n-1}$, because all nonzero matrices have the same probability of being included in $V$.
I'm looking for a concentration bound for $f(V)$, saying that with very high probability over the choice of $V$ we have that $f(V)$ is not much higher than $f(\mathcal{M}_{n,m}(\mathbb{F_q}))$.
Any help is greatly appreciated!