Resuming pretraining from checkpoint

[Lathashree]

Check before submitting issues

• Make sure to pull the latest code, as some issues and bugs have been fixed.
• Due to frequent dependency updates, please ensure you have followed the steps in our Wiki
• I have read the FAQ section AND searched for similar issues and did not find a similar problem or solution
• Third-party plugin issues - e.g., llama.cpp, text-generation-webui, LlamaChat, we recommend checking the corresponding project for solutions
• Model validity check - Be sure to check the model's SHA256.md. If the model is incorrect, we cannot guarantee its performance

Type of Issue

Model training and fine-tuning

Base Model

LLaMA-7B

Operating System

Linux

Describe your issue in detail

I am doing pretraining of the LLAMA 7B model, but due to time limits in the cluster, it stopped. I restarted the program again.

I am getting error - "Exception: Current loss scale already at minimum - cannot decrease scale anymore. Exiting run."

In my ds config, I have min loss scale of 1e-10; my previous training loss stops at 1.092.
I have attached a snapshot of it. I am doing training in fp16, and due to hardware limitations, I can't run on bf16; the job on cluster fails with error.

What can I do further to solve this issue? Any help would be greatly appreciated.

Thanks.

Dependencies (must be provided for code-related issues)

my ds config:

{
    "fp16": {
	"enabled": true,
        "loss_scale": 0,
        "loss_scale_window": 100,
        "initial_scale_power": 16,
        "hysteresis": 2,
        "min_loss_scale": 1e-10
    },
    "bf16": {
	"enabled": false
    },
    "zero_optimization": {
        "stage": 2,
        "offload_optimizer": {
            "device": "cpu",
            "pin_memory": true
        },
	"allgather_partitions": true,
        "allgather_bucket_size": 1e8,
        "overlap_comm": true,
        "reduce_scatter": true,
        "reduce_bucket_size": 1e8,
        "contiguous_gradients": true
    },
    "gradient_accumulation_steps": "auto",
    "gradient_clipping": "auto",
    "steps_per_print": 2000,
    "train_batch_size": "auto",
    "train_micro_batch_size_per_gpu": "auto",
    "wall_clock_breakdown": false
}

Execution logs or screenshots

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your consideration.

[github-actions]

Closing the issue, since no updates observed. Feel free to re-open if you need any further assistance.