The identification of jets containing b-hadrons, b-tagging, plays an important role in many physics analyses in ATLAS. Several different machine learning algorithms have been deployed for the purpose of b-tagging. These tagging algorithms are trained using Monte-Carlo simulation samples, as such their performance in data must be measured. The b-tagging efficiencies have been measured in data using $t\bar{t}$ events in the past and this work presents the measurements in multijet events using data collected by the ATLAS detector at $\sqrt{s}$ = 13 TeV. This offers several key advantages over
the $t\bar{t}$ based calibrations, including a higher precision at low jet $p_T$ and the ability to perform measurements of $\varepsilon_b$ at significantly higher jet $p_T$. Two approaches are applied and for both a profile likelihood fit is performed to extract the number of b-jets in samples passing and failing a given b-tagging requirement. The b-jets yields are then used to determine $\varepsilon_b$ in data and from that scale factors to the efficiency measured in the Monte-Carlo. The two approaches differ primarily in the discriminating variable used in the fit. At low jet $p_T$ the variable muon $p_T^{\text{rel}}$ is used, while for high jet $p_T$ the signed impact parameter significance is used. Both calibrations give measurements of the scale factors as a function of the jet $p_T$ .