transformers

86a6c97c - Pass kwargs to the Mamba2 mixer in Nemotron-H, Falcon-H1 and Mamba2 (#48490)

Pass kwargs to the Mamba2 mixer in Nemotron-H, Falcon-H1 and Mamba2 (#48490) * Pass kwargs to the Nemotron-H Mamba2 mixer NemotronHBlock forwards **kwargs to the attention mixer but dropped them for the linear-attention branch, so kernel kwargs such as seq_idx never reached the mixer. The mixer already splats them into the fused conv1d+scan, the conv fallback and the chunk scan, and Bamba, GraniteMoEHybrid and Zamba2 all pass them through. Co-authored-by: Cursor <cursoragent@cursor.com> * Pass kwargs to the Mamba2 mixer in falcon_h1 and mamba2 too Same gap as nemotron_h: the block accepts **kwargs but drops them at the mixer, so kernel kwargs such as seq_idx never reach the kernels. In these two the model also did not pass kwargs down to the block. Also move the nemotron_h test into the tester and drop the seq_idx pop, which is not needed since the torch fallback kernels take **kwargs. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com>