These _are_ the files that make up a leveldb database. In databases the contents (here: keys and associated values) are usually not exposed as files, but have to be read through some kind of API. In relational databases usually some dialect of SQL is used to formulate queries, in (more recent) key-value store databases there is some kind of API to query a key and retrieve its associated values. WIth caffe, the keys stored in LMDBs and LevelDBs are just 8-digit "integer sequence" strings: 00000000, 00000001, 00000002, ... And the attached values are just serialized versions of the "Datum" protobuf message.
I do not really understand your question, because in Caffe LevelDB/LMDB databases are only used in the "Data" layer to provide input (training data) for the network. Afaik there is no export to LMDB/LevelDB of any kind. The trained network is save in snapshots with the file extension .caffemodel along with the .solverstate, which allows continuing training from the saved state.
Jan