LoRA fine-tuning for fastchat-3b exits with return code = -7

@BabyChouSr is already working on this.

Since Jul 13, 2023.

Hi, I'm fine-tuning a fastchat-3b model with LoRA. The processes are getting killed at the `trainer.train()` step with the following log / error:
```
Loading extension module cpu_adam...
Time to load cpu_adam op: 1.6071059703826904 seconds
Loading extension module cpu_adam...
Time to load cpu_adam op: 1.6841044425964355 seconds
Rank: 1 partition count [2] and sizes[(2359296, False)]
Rank: 0 partition count [2] and sizes[(2359296, False)]
[2023-07-13 10:03:24,669] [INFO] [launch.py:315:sigkill_handler] Killing subprocess 12267
[2023-07-13 10:03:24,802] [INFO] [launch.py:315:sigkill_handler] Killing subprocess 12268
[2023-07-13 10:03:24,802] [ERROR] [launch.py:321:sigkill_handler] ['/opt/conda/bin/python', '-u', sys args...] exits with return code = -7
```

I made the following two changes in code & executing it with below command
1. Commented out the import & use of `replace_llama_attn_with_flash_attn` due to some package issues
2. Updated the below class to suit fastchat model with T5 architecture
```
@dataclass
class LoraArguments:
lora_r: int = 8
lora_alpha: int = 16
lora_dropout: float = 0.05
lora_target_modules: typing.List[str] = field(
#default_factory=lambda: ["q_proj", "v_proj"]
default_factory=lambda: ["q", "v"]
)
lora_weight_path: str = ""
lora_bias: str = "none"
q_lora: bool = False
```
```
deepspeed train_lora.py \
--model_name_or_path checkpoints_fastchat/lmsys/fastchat-t5-3b-v1.0 \
--lora_r 8 \
--lora_alpha 16 \
--lora_dropout 0.05 \
--data_path data/ust/input_data.json \
--bf16 True \
--output_dir data \
--num_train_epochs 4 \
--per_device_train_batch_size 1 \
--per_device_eval_batch_size 0 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 5 \
--save_total_limit 100 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--tf32 True \
--model_max_length 2048 \
--q_lora False \
--deepspeed FastChat/playground/deepspeed_config_s2.json
```

Please suggest some possible issues to look out for & a resolution, if any.