Skip to content
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

Jump to bottom

server: stop generation at n_ctx_train if n_predict is not set #6638

Merged
merged 7 commits into from
Apr 26, 2024
Merged

server: stop generation at n_ctx_train if n_predict is not set #6638

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
6fd5ad5
server: cap n_predict if not set to n_ctx_train
phymbert Apr 12, 2024
558f690
Merge remote-tracking branch 'refs/remotes/origin/master' into hp/ser…
phymbert Apr 19, 2024
5d64ffd
server: fix infinite loop
phymbert Apr 19, 2024
1423fce
server: infinite loop, move in process_token
phymbert Apr 19, 2024
03f458c
minor: spaces
phymbert Apr 19, 2024
0d3eca6
minor: spaces
phymbert Apr 19, 2024
6c257f4
server: include prompt tokens in the EOS limit
phymbert Apr 19, 2024
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
23 changes: 22 additions & 1 deletion examples/server/server.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1208,6 +1208,27 @@ struct server_context {
LOG_VERBOSE("eos token found", {});
}

auto n_ctx_train = llama_n_ctx_train(model);
if (slot.params.n_predict < 1 && slot.ga_n == 1
&& slot.n_prompt_tokens + slot.n_decoded >= n_ctx_train) {
LOG_WARNING("n_predict is not set and self-context extend is disabled."
" Limiting generated tokens to n_ctx_train to avoid EOS-less generation infinite loop", {
{ "id_slot", slot.id },
{ "params.n_predict", slot.params.n_predict },
{ "slot.n_prompt_tokens", slot.n_prompt_tokens },
{ "slot.n_decoded", slot.n_decoded },
{ "slot.n_predict", slot.n_predict },
{ "n_slots", params.n_parallel },
{ "slot.n_ctx", slot.n_ctx },
{ "n_ctx", n_ctx },
{ "n_ctx_train", n_ctx_train },
{ "ga_n", slot.ga_n },
});
slot.truncated = true;
slot.stopped_limit = true;
slot.has_next_token = false; // stop prediction
}

LOG_VERBOSE("next token", {
{"id_slot", slot.id},
{"id_task", slot.id_task},
Expand Down Expand Up @@ -2142,7 +2163,7 @@ struct server_context {
});

// process the created batch of tokens
for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += n_batch) {
for (int32_t i = 0; i < batch.n_tokens; i += n_batch) {
const int32_t n_tokens = std::min(n_batch, batch.n_tokens - i);

for (auto & slot : slots) {
Expand Down
Loading