Single-molecule chromatin fiber sequencing is based on the single-nucleotide resolution identification of DNA N6-methyladenine (m6A) along individual sequencing reads. We present fibertools, a semi-supervised convolutional neural network that permits the fast and accurate identification of both endogenous and exogenous m6A-marked bases using single-molecule long-read sequencing. Fibertools enables highly accurate (>90% precision and recall) m6A identification along multi-kilobase DNA molecules with a ~1,000-fold improvement in speed and the capacity to generalize to new sequencing chemistries.