In a phase 2 clinical trial, the basic target is the proof-of-concept. For a multi-arm dose-ranging study of continuous endpoint, hypothesis tests regarding linear contrast are carried out to confirm efficacy. In this paper, we take both frequentist and Bayesian linear contrast tests into account. For each of them, an adaptive sample size increase based on conditional or predictive power is committed to improve the initial sample size, provided that trial success is promising at the interim analysis. Under the frequentist scheme, the type 1 error of our two-stage design could be preserved via reweighted test statistics. On the other hand, the Bayesian testing framework possesses more flexibility since different prior knowledge could be adopted. Using MCMC and Laplace approximation, our two-stage Bayesian test could be implemented under a wide range of prior settings for both single and multiple testing scenarios. Comparative studies under different simulation settings show that our adaptive procedure is beneficial compared with fixed design. Our method increases the chance of ultimate trial success, especially when the signal is promising but not conclusive at the interim stage.