spec : add DFlash support (#22105)

* spec: add DFlash v2 support

* dflash: support sliding window attention per layer_types

* docs: add dflash section

---------

Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com>
This commit is contained in:
Ruixiang Wang
2026-06-28 16:01:34 +03:00
committed by GitHub
co-authored by Kashif Rasul
parent c1a1c8ee94
commit d1b34251bc
14 changed files with 712 additions and 9 deletions
+2 -1
View File
@@ -169,6 +169,7 @@ enum common_speculative_type {
COMMON_SPECULATIVE_TYPE_DRAFT_SIMPLE, // standalone draft model speculative decoding
COMMON_SPECULATIVE_TYPE_DRAFT_EAGLE3, // Eagle3 speculative decoding
COMMON_SPECULATIVE_TYPE_DRAFT_MTP, // Multi-token prediction
COMMON_SPECULATIVE_TYPE_DRAFT_DFLASH, // DFlash speculative decoding
COMMON_SPECULATIVE_TYPE_NGRAM_SIMPLE, // simple self-speculative decoding based on n-grams
COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K, // self-speculative decoding with n-gram keys only
COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K4V, // self-speculative decoding with n-gram keys and 4 m-gram values
@@ -384,7 +385,7 @@ struct common_params_speculative {
uint32_t need_n_rs_seq() const {
bool needs_rs_seq = std::any_of(types.begin(), types.end(), [&](auto t) {
return t == COMMON_SPECULATIVE_TYPE_DRAFT_MTP || t == COMMON_SPECULATIVE_TYPE_DRAFT_EAGLE3;
return t == COMMON_SPECULATIVE_TYPE_DRAFT_MTP || t == COMMON_SPECULATIVE_TYPE_DRAFT_EAGLE3 || t == COMMON_SPECULATIVE_TYPE_DRAFT_DFLASH;
});
return needs_rs_seq ? draft.n_max : 0u;