AttributeError: 'BiDecoderOnlyEmbedderICLModel' object has no attribute 'config', when tune bge-en-icl with deepspeed zero3

[leyi-123]

Limited by computing resources, I tune bge-en-icl with deepspeed zero3, whose config is as following:

{
    "zero_optimization": {
        "stage": 3,
        "offload_optimizer": {
            "device": "cpu",
            "pin_memory": true
        },
        "offload_param": {
            "device": "cpu",
            "pin_memory": true
        },
        "overlap_comm": true,
        "contiguous_gradients": true,
        "sub_group_size": 1e9,        
        "reduce_bucket_size": "auto",
        "stage3_prefetch_bucket_size": "auto",
        "stage3_param_persistence_threshold": "auto",
        "stage3_max_live_parameters": 1e9,
        "stage3_max_reuse_distance": 1e9,
        "stage3_gather_fp16_weights_on_model_save": true
    },
    "fp16": {
        "enabled": "auto",
        "loss_scale": 0,
        "initial_scale_power": 10,
        "loss_scale_window": 1000,
        "hysteresis": 2,
        "min_loss_scale": 1
    },
    "bf16": {
        "enabled": "auto",
        "loss_scale": 0,
        "initial_scale_power": 10,
        "loss_scale_window": 1000,
        "hysteresis": 2,
        "min_loss_scale": 1
    },
    "optimizer": {
        "type": "AdamW",
        "params": {
            "lr": "auto",
            "betas": "auto",
            "eps": "auto",
            "weight_decay": "auto",
            "torch_adam": true
        }
    },
    "scheduler": {
        "type": "WarmupDecayLR",
        "params": {
            "warmup_min_lr": "auto",
            "warmup_max_lr": "auto",
            "warmup_num_steps": "auto",
            "total_num_steps": "auto"
        }
    },

    "gradient_accumulation_steps": "auto",
    "gradient_clipping": "auto",
    "steps_per_print": 1000,
    "train_batch_size": "auto",
    "train_micro_batch_size_per_gpu": "auto",
    "wall_clock_breakdown": false
}

However, at the beginning of the tuning process, I meet the following error message:

[rank0]: Traceback (most recent call last): 
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/runpy.py", line 196, in _run_module_as_main
[rank0]:     return _run_code(code, main_globals, None,
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/runpy.py", line 86, in _run_code
[rank0]:     exec(code, run_globals)
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/FlagEmbedding/finetune/embedder/decoder_only/icl/__main__.py", line 26, in <module>
[rank0]:     runner.run()
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/FlagEmbedding/finetune/embedder/decoder_only/icl/runner.py", line 157, in run
[rank0]:     self.trainer.train(resume_from_checkpoint=self.training_args.resume_from_checkpoint)
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/transformers/trainer.py", line 2164, in train
[rank0]:     return inner_training_loop(
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/transformers/trainer.py", line 2262, in _inner_training_loop
[rank0]:     self.optimizer, self.lr_scheduler = deepspeed_init(self, num_training_steps=max_steps)
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/transformers/integrations/deepspeed.py", line 398, in deepspeed_init
[rank0]:     hf_deepspeed_config.trainer_config_finalize(args, model, num_training_steps)
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/transformers/integrations/deepspeed.py", line 226, in trainer_config_finalize
[rank0]:     if hasattr(model.config, "hidden_size"):
[rank0]:   File "/data//miniconda3/envs/bge/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1729, in __getattr__ 
[rank0]:     raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'") 
[rank0]: AttributeError: 'BiDecoderOnlyEmbedderICLModel' object has no attribute 'config' 

My torch version is 2.4.0+cu118, transformers version is 4.47.1 and deepspeed version is 0.16.7. I want to know how to tune bge-en-icl (based on mistral-7B) with deepspeed zero3. Thanks!

[545999961]

Sorry, our current training is not compatible with DeepSpeed ZeRO3. We recommend using a lower stage.