distributed_type: DEEPSPEED deepspeed_config: deepspeed_config_file: tests/trainer/distributed/scripts/ds_config_zero2.json num_processes: 2 parallelism_config: parallelism_config_sp_size: 2 parallelism_config_sp_backend: deepspeed parallelism_config_sp_seq_length_is_variable: true parallelism_config_sp_attn_implementation: sdpa