flashattnvarlen support tree attention

[efsotr]

flashattnvarlen support tree attention (doesn't support tree attention + alibi / sliding window)

[v-lmn]

Thank you very much, I want to understand some questions.Can you explain some of your test case,I don't understand what's the meaning of "~create_mask(max_seqlen, half_prompts_lens)" and "mask" and "start_pos_id" and "end_pos_id". Can you give me your personal email?

[v-lmn]

start_pos_id = create_pos_id(lens, [0, 1, 2])
end_pos_id = create_pos_id(lens, [2, 1, 2])
why [0, 1, 2] and [2, 1, 2]

[v-lmn]

hi

[v-lmn]

can you help