Skip to content

to benchmark chunked long context prefill #9779

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

Open
wants to merge 1 commit into
base: gh/kimishpatel/163/base
Choose a base branch
from
Open

to benchmark chunked long context prefill #9779

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
26 changes: 21 additions & 5 deletions examples/models/llama/runner/runner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,7 @@

#include <executorch/examples/models/llama/runner/runner.h>

#include <algorithm>
#include <ctime>

#include <executorch/extension/llm/runner/util.h>
Expand Down Expand Up @@ -221,11 +222,11 @@ Error Runner::generate(

ET_CHECK_MSG(num_prompt_tokens >= 1, "Expected at least 1 prompt token");
ET_CHECK_MSG(
num_prompt_tokens < metadata_.at(kMaxSeqLen),
num_prompt_tokens < metadata_.at(kMaxContextLen),
"num_prompt_tokens %d >= max_seq_len_ %" PRId64
", Max seq length exceeded - please increase max seq len value in .../llama2/model.py",
num_prompt_tokens,
metadata_.at(kMaxSeqLen));
metadata_.at(kMaxContextLen));
ET_CHECK_MSG(
num_prompt_tokens < seq_len,
"num_prompt_tokens %d >= seq_len %d, Sequence length exceeded - please increase the seq_len value passed to generate()",
Expand All @@ -241,11 +242,26 @@ Error Runner::generate(
wrapped_callback(prompt);
}
int64_t pos = 0;
auto prefill_res = text_prefiller_->prefill(prompt_tokens, pos);
uint64_t cur_token;
int max_seq_len = metadata_.at(kMaxSeqLen) -
1; // -1 because for some reason tracing results in this upperbound
int num_tokens_to_process = 0;
while (num_tokens_to_process < num_prompt_tokens) {
auto num_tokens_to_prefill_with =
std::min(num_prompt_tokens - num_tokens_to_process, max_seq_len);
std::vector<uint64_t> prompt_tokens_to_process(num_tokens_to_prefill_with);
std::copy(
prompt_tokens.begin() + num_tokens_to_process,
prompt_tokens.begin() + num_tokens_to_process + num_tokens_to_prefill_with,
prompt_tokens_to_process.begin());
auto prefill_res =
text_prefiller_->prefill(prompt_tokens_to_process, pos);
ET_CHECK_OK_OR_RETURN_ERROR(prefill_res.error());
cur_token = prefill_res.get();
num_tokens_to_process += num_tokens_to_prefill_with;
}
stats_.first_token_ms = llm::time_in_ms();
stats_.prompt_eval_end_ms = llm::time_in_ms();
ET_CHECK_OK_OR_RETURN_ERROR(prefill_res.error());
uint64_t cur_token = prefill_res.get();

// print the first token from prefill. No prev_token so use cur_token for it.
wrapped_callback(
Expand Down
Loading