Skip to content

Commit

Permalink
[fine_tuning]: debug accelerate launch
Browse files Browse the repository at this point in the history
  • Loading branch information
albertoperdomo2 committed Jan 2, 2025
1 parent 0c40412 commit 9bf2aff
Show file tree
Hide file tree
Showing 2 changed files with 2 additions and 7 deletions.
1 change: 1 addition & 0 deletions projects/fine_tuning/testing/config.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -99,6 +99,7 @@ ci_presets:

tests.fine_tuning.test_settings.hyper_parameters.gradient_accumulation_steps: 1
tests.fine_tuning.test_settings.hyper_parameters.per_device_train_batch_size: 4
tests.fine_tuning.test_settings.hyper_parameters.log_level: "debug"
tests.fine_tuning.test_settings.hyper_parameters.peft_method: "none"
tests.fine_tuning.test_settings.hyper_parameters.max_seq_length: 1024
tests.fine_tuning.test_settings.hyper_parameters.use_flash_attn: true
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -16,13 +16,7 @@ if [[ $WORLD_SIZE == 1 ]]; then
else
echo "Running with a $NUM_GPUS GPUs"
fi
time python -m torch.distributed.run \
--node_rank "$RANK" \
--nnodes "$WORLD_SIZE" \
--nproc_per_node "$(nvidia-smi --query-gpu=gpu_name --format=csv,noheader | wc -l) " \
--master_addr "$MASTER_ADDR" \
--master_port "$MASTER_PORT" \
launch_training.py
time python /app/accelerate_launch.py
exit 0
fi
echo "Running on $WORLD_SIZE machines with $NUM_GPUS GPUs each."
Expand Down

0 comments on commit 9bf2aff

Please sign in to comment.