Multiple industrial processes involve gas-liquid flows characterized by a wide range of spatial and temporal scales. Simulating such flows remains a major challenge nowadays, as the computational cost associated with Direct Numerical Simulation still makes it unaffordable. For such configurations, an interesting alternative to DNS is the use of multi-scale approaches. In the latter, large enough bubbles are fully resolved and may deform over time, while smaller bubbles are modeled as a dispersed phase using subgrid scale models. The interfacial momentum transfer terms are then tailored to the local flow configuration. The closure models still involved in these approaches and the influence of the cut-off length separating the resolved and modeled bubbles definitely need to be validated against detailed experiments. In order to assess the validity of these models, we present a one-to-one comparison between experiments performed in a simple configuration, namely the emptying of a water bottle, and numerical simulations using the aforementioned approach. The results are found to reliably reproduce the genesis of the oscillation mechanism, which is governed by the bubble formation at the bottle neck. The multi-scale model also qualitatively reproduces the fragmentation process of large bubbles during their rise in the water column. However local experimental data are required to assess more quantitatively these results.