Deepseek v3.2 no supported?

[FriskFreeman]

I've noticed that there's no implementation of the new architecture feature (DSA, sparse attn for Deepseek v3.2) in the main simulation procedure.
Though there is "index_topk" in model_config (but without index_head_dim, index_n_heads, etc...). Also a kernel_sim program for DSA (sparse_mla_fp8.py) and some csv result files. But it seems unrelated to the main and model code.

Is the DSA feature in roadmap?

[MtFitzRoy]

@FriskFreeman Yes, we are trying to simulate the KV Cache offloading for DeepSeek-V3.2, have got some results, but not ready for open source.