Update model card: codeset_v4_seq2048 (37,618 rows, multi-language, 1 epoch) pass@1=0.004106776180698152

codeset_v4-seq2048 (37,618-row multi-language corpus, 1 epoch) trained directly from pristine Minitron-4B-Base, single-file MXFP8, byte-identical to trainer master

Update model card: codeset_v4_expanded (37,618 rows, multi-language, 1 epoch) pass@1=0.002053388090349076

codeset_v4-expanded (37,618-row multi-language corpus, 1 epoch) trained directly from pristine Minitron-4B-Base, single-file MXFP8, byte-identical to trainer master

Update model card with real pass@1=0.0

codeset_v3 SFT (plain text) trained directly from pristine Minitron-4B-Base (Dolly skipped), single-file MXFP8, byte-identical to trainer master

Initial model card: training in progress

initial commit