Skip to content

server : keep speculative recurrent-state checkpoints on-device #28118

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

Draft
vahpetr wants to merge 1 commit into
base: master
Choose a base branch
from
+8 −8
Draft

server : keep speculative recurrent-state checkpoints on-device #28118

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
16 changes: 8 additions & 8 deletions tools/server/server-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2993,7 +2993,7 @@ struct server_context_impl {
llama_memory_seq_pos_max(llama_get_memory(ctx_tgt), slot.id));

if (use_ckpt_dft) {
slot.spec_ckpt.update_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
slot.spec_ckpt.update_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
}

slot.spec_prompt = slot.prompt.tokens.get_text_tokens();
Expand Down Expand Up @@ -3032,7 +3032,7 @@ struct server_context_impl {

if (ctx_dft) {
if (use_ckpt_dft) {
ckpt.load_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
ckpt.load_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
}

if (!llama_memory_seq_rm(llama_get_memory(ctx_dft), slot.id, ckpt.pos_max + 1, -1)) {
Expand All @@ -3051,7 +3051,7 @@ struct server_context_impl {
if (use_ckpt_tgt) {
//const int64_t t_start = ggml_time_us();

ckpt.update_tgt(ctx_tgt, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
ckpt.update_tgt(ctx_tgt, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);

//const int64_t t_total = ggml_time_us() - t_start;
//printf("checkpoint total: %f ms\n", t_total / 1000.0);
Expand All @@ -3063,7 +3063,7 @@ struct server_context_impl {
}

if (use_ckpt_dft) {
ckpt.update_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
ckpt.update_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
}
}
});
Expand Down Expand Up @@ -3339,8 +3339,8 @@ struct server_context_impl {

if (!do_reset) {
// restore the context checkpoint
it->load_tgt(ctx_tgt, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
it->load_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
it->load_tgt(ctx_tgt, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
it->load_dft(ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
// restore the draft's speculative state
common_speculative_set_state(spec.get(), slot.id, it->data_spec);

Expand Down Expand Up @@ -3915,10 +3915,10 @@ struct server_context_impl {

SLT_DBG(slot, "restoring speculative checkpoint (pos_min = %d, pos_max = %d, size = %zu)\n", ckpt.pos_min, ckpt.pos_max, ckpt.size());

ckpt.load_tgt(slot.ctx_tgt, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
ckpt.load_tgt(slot.ctx_tgt, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);

if (slot.ctx_dft) {
ckpt.load_dft(slot.ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY);
ckpt.load_dft(slot.ctx_dft, slot.id, LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY | LLAMA_STATE_SEQ_FLAGS_ON_DEVICE);
}

slot.mem.seq_rm(slot.id, ckpt.pos_max + 1, -1);
Expand Down