Skip to content

gemma : allow offloading the output tensor #5646

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
Feb 21, 2024
Merged

gemma : allow offloading the output tensor #5646

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: 3 additions & 1 deletion llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4394,6 +4394,8 @@ static bool llm_load_tensors(

// output
model.output_norm = ml.create_tensor(ctx_output, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
model.output = ml.create_tensor(ctx_output, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}); // same as tok_embd, duplicated to allow offloading
ml.n_created--; // artificial tensor

const int64_t n_ff = hparams.n_ff;
const int64_t n_embd_head_k = hparams.n_embd_head_k;
Expand Down Expand Up @@ -7525,7 +7527,7 @@ struct llm_build_context {
cb(cur, "result_norm", -1);

// lm_head
cur = ggml_mul_mat(ctx0, model.tok_embd, cur);
cur = ggml_mul_mat(ctx0, model.output, cur);
cb(cur, "result_output", -1);

ggml_build_forward_expand(gf, cur);
Expand Down