Skip to content

Honor MAGPIE_EVAL_TOKENIZED_REQUESTS in remote-compat lm_eval model_args #74

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
weilei0120 merged 1 commit into from
Aug 5, 2026
+8 −1
Merged

Honor MAGPIE_EVAL_TOKENIZED_REQUESTS in remote-compat lm_eval model_args #74

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
9 changes: 8 additions & 1 deletion Magpie/scripts/benchmark/magpie_bench_remote_compat.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,8 +136,15 @@ magpie_run_eval_remote_direct() {
# base_url ends in /v1/completions; tokenizer_backend=huggingface uses
# the local hub tokenizer (model path/id) so we don't pay a server-side
# tokenization roundtrip.
#
# MAGPIE_EVAL_TOKENIZED_REQUESTS (optional) controls the prompt wire format.
# Unset => lm_eval's default (token-id-array prompts), which a direct sglang
# server accepts. A PD-disaggregated sglang_router's /v1/completions only
# accepts StringOrArray and rejects token-id arrays with HTTP 422, collapsing
# the accuracy eval; set MAGPIE_EVAL_TOKENIZED_REQUESTS=false there to send
# string prompts instead. Absent env => byte-for-byte the previous behaviour.
local base_url="${BENCHMARK_BASE_URL%/}/v1/completions"
local model_args="model=${MODEL},base_url=${base_url},num_concurrent=${conc},tokenizer_backend=huggingface,trust_remote_code=true"
local model_args="model=${MODEL},base_url=${base_url},num_concurrent=${conc},tokenizer_backend=huggingface,trust_remote_code=true${MAGPIE_EVAL_TOKENIZED_REQUESTS:+,tokenized_requests=${MAGPIE_EVAL_TOKENIZED_REQUESTS}}"
local -a cmd=(
"$py" -m lm_eval
--model local-completions
Expand Down