Skip to content

feat: support resuming existing W&B runs #102

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
Zhuang-Zhuang-Liu wants to merge 3 commits into
base: main
Choose a base branch
from
Open

feat: support resuming existing W&B runs #102

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
20ded25
Add wandb-resume argument to argument parser
Zhuang-Zhuang-Liu Apr 24, 2026
beefbb5
Implement W&B run resumption feature
Zhuang-Zhuang-Liu Apr 24, 2026
7889105
Move W&B resume options into WANDB_ARG
Zhuang-Zhuang-Liu Apr 24, 2026
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
1 change: 1 addition & 0 deletions slime/slime/utils/arguments.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1179,6 +1179,7 @@ def add_wandb_arguments(parser):
help="Whether to turn on passrate logging, which will log the pass@n of the responses in the rollout.",
)
parser.add_argument("--wandb-run-id", type=str, default=None)
parser.add_argument("--wandb-resume", type=str, default="allow")
return parser

# tensorboard
Expand Down
7 changes: 7 additions & 0 deletions slime/slime/utils/wandb_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,6 +58,13 @@ def init_wandb_primary(args):
"config": _compute_config_for_logging(args),
}

# When resuming training, continue logging to the same W&B run if a run id is provided
wandb_run_id = getattr(args, "wandb_run_id", None)
if wandb_run_id is not None:
init_kwargs["id"] = wandb_run_id
init_kwargs["resume"] = args.wandb_resume
logger.info("Resuming W&B run %s with resume=%s", wandb_run_id, args.wandb_resume)

# Configure settings based on offline/online mode
if offline:
init_kwargs["settings"] = wandb.Settings(mode="offline")
Expand Down
8 changes: 5 additions & 3 deletions toolcall-rl/retool_qwen3_4b_rl.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,8 +51,7 @@ HF_CKPT=${HF_CKPT:-/data_storage/wyj/systems/huggingface/hub/qwen3-4b-retool-sft
REF_LOAD=${REF_LOAD:-/data_storage/wyj/systems/huggingface/hub/qwen3-4b-retool-sft_torch_dist}
SAVE_CKPT=${SAVE_CKPT:-/data_storage/wyj/OpenClaw-RL/ckpt/qwen3-4b-retool-rl/}
RESUME_LOAD=${RESUME_LOAD:-${SAVE_CKPT}}
# Use the existing run id to continue plotting on the same W&B curve.
#WANDB_RESUME=${WANDB_RESUME:-must}


CKPT_ARGS=(
--hf-checkpoint ${HF_CKPT}
Expand Down Expand Up @@ -136,6 +135,9 @@ WANDB_ARGS=(
--wandb-project slime_retool
--wandb-group qwen3-4B-rl_retool
--wandb-key ${WANDB_KEY}
# Use the existing run id to continue plotting on the same W&B curve.
# --wandb-run-id <existing_run_id>
# --wandb-resume must
)

SGLANG_ARGS=(
Expand Down Expand Up @@ -192,4 +194,4 @@ ray job submit --address="http://127.0.0.1:8265" \
${EVAL_ARGS[@]} \
${SGLANG_ARGS[@]} \
${MISC_ARGS[@]} \
${CUSTOM_ARGS[@]}
${CUSTOM_ARGS[@]}