nemo_automodel.components.models.glm_moe_dsa.kernels.tilelang_sparse_mla_fwd
nemo_automodel.components.models.glm_moe_dsa.kernels.tilelang_sparse_mla_fwd
Module Contents
Functions
| Name | Description |
|---|---|
sparse_mla_fwd | - |
sparse_mla_fwd_interface | - |
API
nemo_automodel.components.models.glm_moe_dsa.kernels.tilelang_sparse_mla_fwd.sparse_mla_fwd( heads, dim, tail_dim, topk, kv_group = 1, sm_scale = None, is_causal = True, CP0 = True, block_I = 64, num_stages = 2, threads = 256 )
nemo_automodel.components.models.glm_moe_dsa.kernels.tilelang_sparse_mla_fwd.sparse_mla_fwd_interface( q, kv, indices, sm_scale = None, return_p_sum: bool = False, d_v = 512, block_I = 64, num_stages = 2, threads = 256 )