Skip to content

llama : handle KV shift for recurrent models #10402

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 21, 2024
Merged

llama : handle KV shift for recurrent models #10402

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 5 additions & 5 deletions src/llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18211,13 +18211,13 @@ static void llama_kv_cache_defrag_internal(struct llama_context & lctx) {
static void llama_kv_cache_update_internal(struct llama_context & lctx) {
bool need_reserve = false;

// apply K-shift if needed
if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE && lctx.kv_self.has_shift) {
if (lctx.kv_self.has_shift) {
if (!llama_kv_cache_can_shift(&lctx)) {
GGML_ABORT("Deepseek2 does not support K-shift");
GGML_ABORT("The current context does not support K-shift");
}

{
// apply K-shift if needed
if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE) {
ggml_backend_sched_reset(lctx.sched.get());

ggml_cgraph * gf = llama_build_graph_k_shift(lctx);
Expand Down Expand Up @@ -20463,7 +20463,7 @@ void llama_kv_cache_update(struct llama_context * ctx) {
}

bool llama_kv_cache_can_shift(struct llama_context * ctx) {
return ctx->model.arch != LLM_ARCH_DEEPSEEK2; // not supported due to MLA
return !ctx->kv_self.recurrent && ctx->model.arch != LLM_ARCH_DEEPSEEK2; // not supported due to MLA
}

// deprecated
Expand Down
Loading