I have a 20 samples with mutations ranging from 50 to 144 to extract signatures. I don't find an inflection point to determine rank. It seems like the more ranks I use the "better" fit. i suppose with the small number of samples i have there really can't be more than 2 or 3 signatures. Also as i look at the samples contributing to the signatures, some have multiple similar signatures. 

What you most likely see here is an effect of the mutation data itself. As you wrote, each of your samples has 50 to 114 mutations. This is a very low number of variants to infer general characteristics from. Consider that we have 96 possible motifs which means that each sample on average has around 1 variant per motif. You can have a look at this question for a detailed discussion of such cases: SomaticSignatures low mutation rates. Instead of trying to identify signatures on a per-sample base, you could try to define another relevant grouping and aggregate the variants by this groups. The case study covered in the vignette is using this approach, where signatures are estimated at the level of studies.