Analyze

What do we have

Now that we have finished tuning, let’s see what we have. Every 1000 steps, a checkpoint is created. The last checkpoint represents the fully tuned model. For example, the folder checkpoint-15000 contains the final model. The folder structure looks like this:

project fine-tunning
└─── checkpoint-15000
│   │   model.safetensors
│   │   optimizer.bin
│   │   pytorch_lora_weights.safetensors
│   │   random_states_0.pkl
│   │   scheduler.bin
│   └───

The key files in this directory are those with the .safetensors extension:

  • model.safetensors
  • pytorch_lora_weights.safetensors
model.safetensorstorch_lora_weights.safetensors
sizeas size the base modelsmall
descriptionthe original model with tunned model combined inonly the tunned model