Hi @Linda_Abenaim,
If you look at the example from the tutorial I linked above, it shows that the ICU table and representative sequence file were the actual user created feature-table and rep-seqs. So for example if you ran DADA2, then you would have both these files generated at the end of that.
With non-V4 data, based on the example in the tutorial:
$ qiime greengenes2 non-v4-16s \
--i-table icu.biom.qza \ #your own biom table
--i-sequences icu.fna.qza \ #your own rep-seqs file
--i-backbone 2022.10.backbone.full-length.fna.qza \ #you download this
--o-mapped-table icu.gg2.biom.qza \ #this is new gg2-filtered feature-table you use downstream
--o-representatives icu.gg2.fna.qza #this is new gg2-fitlered rep-seqs file used downstream.
In the background, this command is using OTU clustering on your V3-V4 reads and then it inserts those into a background tree.
Once you have these files, you can simply get your taxonomy file
$ qiime greengenes2 taxonomy-from-table \
--i-reference-taxonomy 2022.10.taxonomy.asv.nwk.qza \ #download from gg2
--i-table icu.gg2.biom.qza \ #table from last step
--o-classification icu.gg2.taxonomy.qza #new taxonomy file to use downstream
So, technically you don't need to do any of the Naive Bayes classification steps, and in fact if you look at the bottom of the tutorial in this section it recommends against using it. But, if you did want to use it anyways for some reason, you can download the full-length pre-trained classifier they provide in that page and follow the usual steps you would for your NB workflow.
Hope this helps!