Try limiting your calibration range or reducing the injection volume.

Are the hydrogen concentrations much lower than those of carbon dioxide in your standards? Also, are you using split injection?

You have likely exceeded the linear dynamic range for the carbon dioxide in your standards, but not for the hydrogen (thus my question about the relative concentrations of each gas). The TCD response is based on the varied thermal conductivities of different gases and once you get a high concentration of an analyte gas in the detector for a brief period (with injection of a high-concentration sample) further increases in concentration for subsequent injections do not produce increased detector response as much compared as similar increases would cause at lower concentrations. My guess is that you have a curve that rises asymptotically towards a maximum.