Skip to content

examples : allow extracting embeddings from decoder contexts #13797

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
May 26, 2025
+10 −16
Merged

examples : allow extracting embeddings from decoder contexts #13797

Show file tree
Hide file tree
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
4 changes: 2 additions & 2 deletions examples/embedding/embedding.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,8 +41,8 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu

// run model
LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);
if (llama_encode(ctx, batch) < 0) {
LOG_ERR("%s : failed to encode\n", __func__);
if (llama_decode(ctx, batch) < 0) {
LOG_ERR("%s : failed to process\n", __func__);
}

for (int i = 0; i < batch.n_tokens; i++) {
Expand Down
12 changes: 6 additions & 6 deletions examples/retrieval/retrieval.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,14 +81,14 @@ static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & toke
}
}

static void batch_encode(llama_context * ctx, llama_batch & batch, float * output, int n_seq, int n_embd) {
static void batch_process(llama_context * ctx, llama_batch & batch, float * output, int n_seq, int n_embd) {
// clear previous kv_cache values (irrelevant for embeddings)
llama_kv_self_clear(ctx);

// run model
LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);
if (llama_encode(ctx, batch) < 0) {
LOG_ERR("%s : failed to encode\n", __func__);
if (llama_decode(ctx, batch) < 0) {
LOG_ERR("%s : failed to process\n", __func__);
}

for (int i = 0; i < batch.n_tokens; i++) {
Expand Down Expand Up @@ -233,7 +233,7 @@ int main(int argc, char ** argv) {
// encode if at capacity
if (batch.n_tokens + n_toks > n_batch) {
float * out = emb + p * n_embd;
batch_encode(ctx, batch, out, s, n_embd);
batch_process(ctx, batch, out, s, n_embd);
common_batch_clear(batch);
p += s;
s = 0;
Expand All @@ -246,7 +246,7 @@ int main(int argc, char ** argv) {

// final batch
float * out = emb + p * n_embd;
batch_encode(ctx, batch, out, s, n_embd);
batch_process(ctx, batch, out, s, n_embd);

// save embeddings to chunks
for (int i = 0; i < n_chunks; i++) {
Expand All @@ -267,7 +267,7 @@ int main(int argc, char ** argv) {
batch_add_seq(query_batch, query_tokens, 0);

std::vector<float> query_emb(n_embd, 0);
batch_encode(ctx, query_batch, query_emb.data(), 1, n_embd);
batch_process(ctx, query_batch, query_emb.data(), 1, n_embd);

common_batch_clear(query_batch);

Expand Down
2 changes: 1 addition & 1 deletion src/llama-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -852,7 +852,7 @@ int llama_context::encode(llama_batch & inp_batch) {

int llama_context::decode(llama_batch & inp_batch) {
if (!memory) {
LLAMA_LOG_WARN("%s: cannot decode batches with this context (use llama_encode() instead)\n", __func__);
LLAMA_LOG_DEBUG("%s: cannot decode batches with this context (calling encode() instead)\n", __func__);
return encode(inp_batch);
}

Expand Down
8 changes: 1 addition & 7 deletions tools/server/server.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3394,13 +3394,7 @@ struct server_context {
batch.logits + i,
};

int ret = 0;

if (do_encode) {
ret = llama_encode(ctx, batch_view);
} else {
ret = llama_decode(ctx, batch_view);
}
const int ret = llama_decode(ctx, batch_view);

metrics.on_decoded(slots);

Expand Down
Loading