[Question]: Some questions on the code

Describe the issue

Hi, thanks for your great work!

I have some questions about your code:

1. In search_pattern, the search space is different from Table.6 in the paper? And the search space is generated base on the fixed FLOPs?

2. In the following code snippets of search_pattern, every best_ty is reassigned to "vertical_and_slash", so only vertical_slash_sparse_attention will be called? Besides, for best_ty == "block_sparse", where does the magic number 1000 and 6096 come from?

if best_ty == "stream_llm":
    best_ty = "vertical_and_slash"
if best_ty == "block_sparse":
    best_ty, best_v, best_s = "vertical_and_slash", 1000, 6096

3. In the paper, you mentioned one sample is enough to capture the pattern for different prompts with different lengths, any support for that?

we use only one sample as our validation set from KV retrieval synthetic data with 30k token inputs, which exhibits strong generalization and stability across different lengths and domains.

Besides, I think I might find a bug when I run examples/run_hf_streaming.sh with Qwen2-7B-Instruct. In apply_rotary_embed_single, you need to enable position_ids since q_len == 1 during decoding stage. I change the code into cos = cos[position_ids] sin = sin[position_ids] and it's fine now.

A bit wordy... Looking forward to your reply!

### Tasks