Skip to content

llama : skip token bounds check when evaluating embeddings #9437

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
Sep 11, 2024
Merged

llama : skip token bounds check when evaluating embeddings #9437

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
32 changes: 18 additions & 14 deletions src/llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16076,19 +16076,21 @@ static int llama_decode_internal(
return -1;
}

for (uint32_t i = 0; i < n_tokens_all; ++i) {
if (batch_all.token[i] < 0 || (uint32_t)batch_all.token[i] >= lctx.model.vocab.n_vocab) {
LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch_all.token[i]);
return -1;
}
}

const auto & model = lctx.model;
const auto & hparams = model.hparams;
const auto & cparams = lctx.cparams;

GGML_ASSERT((!batch_all.token && batch_all.embd) || (batch_all.token && !batch_all.embd)); // NOLINT

if (batch_all.token) {
for (uint32_t i = 0; i < n_tokens_all; ++i) {
if (batch_all.token[i] < 0 || (uint32_t)batch_all.token[i] >= model.vocab.n_vocab) {
LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch_all.token[i]);
return -1;
}
}
}

GGML_ASSERT(n_tokens_all <= cparams.n_batch);

GGML_ASSERT((cparams.causal_attn || cparams.n_ubatch >= n_tokens_all) && "non-causal attention requires n_ubatch >= n_tokens");
Expand Down Expand Up @@ -16375,19 +16377,21 @@ static int llama_encode_internal(
return -1;
}

for (uint32_t i = 0; i < n_tokens; ++i) {
if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= lctx.model.vocab.n_vocab) {
LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch.token[i]);
return -1;
}
}

const auto & model = lctx.model;
const auto & hparams = model.hparams;
const auto & cparams = lctx.cparams;

GGML_ASSERT((!batch.token && batch.embd) || (batch.token && !batch.embd)); // NOLINT

if (batch.token) {
for (uint32_t i = 0; i < n_tokens; ++i) {
if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= model.vocab.n_vocab) {
LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch.token[i]);
return -1;
}
}
}

// micro-batching is not possible for non-causal encoding, so we process the batch in a single shot
GGML_ASSERT(cparams.n_ubatch >= n_tokens && "encoder requires n_ubatch >= n_tokens");

Expand Down
Loading