diff --git a/.github/workflows/test_torchtitan.yml b/.github/workflows/test_torchtitan.yml index 9d3c3f9..ea0adc4 100644 --- a/.github/workflows/test_torchtitan.yml +++ b/.github/workflows/test_torchtitan.yml @@ -59,3 +59,14 @@ jobs: --parallelism.data_parallel_shard_degree 4 \ --parallelism.expert_parallel_degree 4 \ --job.custom_config_module=torchtitan.experiments.autoparallel.job_config + + # TorchTitan DSv3 SimpleFSDP with auto_bucketing passes + NGPU=4 CONFIG_FILE=./torchtitan/models/deepseek_v3/train_configs/debug_model.toml ./run_train.sh \ + --model.name simple_fsdp.deepseek_v3 \ + --parallelism.tensor_parallel_degree 1 \ + --parallelism.expert_parallel_degree 4 \ + --job.custom_config_module=torchtitan.experiments.simple_fsdp.job_config \ + --compile.graph_passes auto_bucketing \ + --activation_checkpoint.mode "none" \ + --compile.backend inductor \ + --compile.enable