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.safetensorspytorch_lora_weights.safetensors
| model.safetensors | torch_lora_weights.safetensors | |
|---|---|---|
| size | as size the base model | small |
| description | the original model with tunned model combined in | only the tunned model |