Skip to content

fix(deepswe): Configure default EOS_TOKENS and plumb into rollout launchers #2314

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
andytwigg wants to merge 1 commit into
base: atwigg/mlperf
Choose a base branch
from
+8 −1
Open

fix(deepswe): Configure default EOS_TOKENS and plumb into rollout launchers #2314

Show file tree
Hide file tree
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: 4 additions & 0 deletions tunix/experimental/examples/deepswe_dist/k8s_launcher.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,9 @@ export USE_LORA=${USE_LORA:-0}
export TEMPERATURE=${TEMPERATURE:-1.0}
export TOP_P=${TOP_P:-1.0}
export TOP_K=${TOP_K:--1}
# Model-specific EOS token IDs (comma-separated), fetched from HuggingFace
# `generation_config.json`. Empty string falls back to the tokenizer's default EOS token.
export EOS_TOKENS=${EOS_TOKENS-'151645,151643'}
export DEBUG=${DEBUG:-0}
export USE_ROLLOUT_LOGPS=${USE_ROLLOUT_LOGPS:-true}
export SAMPLER=${SAMPLER:-inprocess_vllm}
Expand Down Expand Up @@ -608,6 +611,7 @@ if cfg:
--mesh_tp=${ROLLOUT_MESH_TP} \
--max_prompt_length=${MAX_PROMPT_LENGTH} \
--max_response_length=${MAX_RESPONSE_LENGTH} \
${EOS_TOKENS:+--eos_tokens=\"${EOS_TOKENS}\"} \
--sampler=${SAMPLER} \
--lora_rank=${LORA_RANK} \
--lora_alpha=${LORA_ALPHA} \
Expand Down
5 changes: 4 additions & 1 deletion tunix/experimental/examples/deepswe_dist/launcher.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,9 @@ LORA_ALPHA=${LORA_ALPHA:-64.0}
TEMPERATURE=${TEMPERATURE:-1.0}
TOP_P=${TOP_P:-1.0}
TOP_K=${TOP_K:--1}
EOS_TOKENS=${EOS_TOKENS-}
# Model-specific EOS token IDs (comma-separated), fetched from HuggingFace
# `generation_config.json`. Empty string falls back to the tokenizer's default EOS token.
EOS_TOKENS=${EOS_TOKENS-'151645,151643'}
MAXTEXT_ATTENTION=${MAXTEXT_ATTENTION:-}
# DEBUG=1 passes --debug to the runner, which logs full sampler responses.
DEBUG=${DEBUG:-0}
Expand Down Expand Up @@ -284,6 +286,7 @@ echo " beta: ${BETA}"
echo " epsilon: ${EPSILON}"
echo " sampler: ${SAMPLER}"
echo " weight sync: ${WEIGHT_SYNC_MODE}"
echo " eos tokens: ${EOS_TOKENS:-<tokenizer default>}"
echo " trainer backend:${TRAINER_BACKEND}"
echo " maxtext model: ${MAXTEXT_MODEL_NAME:-<unset>}"
echo " maxtext ckpt: ${MAXTEXT_CKPT:-<unset>}"
Expand Down
Loading