### Run GALLa First Stage Pretraining

Source: https://github.com/codefuse-ai/galla/blob/main/README.md

Execute the first stage of GALLa training, which focuses on pretraining the model to recover source code from graph data. This command uses `accelerate` for distributed training, referencing `accelerate_ds_config.yaml` for environment setup and `configs/config_pretrain.json` for training parameters.

```Bash
accelerate launch --config_file accelerate_ds_config.yaml run.py --train_config configs/config_pretrain.json
```

--------------------------------

### Run GALLa Second Stage Instruction Finetuning

Source: https://github.com/codefuse-ai/galla/blob/main/README.md

Execute the second stage of GALLa training, which involves instruction finetuning. This command uses `accelerate` for distributed training, referencing `accelerate_ds_config.yaml` for environment setup and `configs/config_instruction.json` for training parameters. The `checkpoint` in the second stage's training config should be the path to the checkpoint saved in the first stage's training.

```Bash
accelerate launch --config_file accelerate_ds_config.yaml run.py --train_config configs/config_instruction.json
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.