| ## Installation Instructions | |
| As a pre-requisite, make sure you have [ducttape](https://github.com/CoderPat/ducttape) and [(mini)conda](https://docs.conda.io/en/latest/miniconda.html) installed. | |
| First, clone this repository. | |
| Then, to create a new conda environment with all the necessary dependencies, run the following command: | |
| ```bash | |
| export CONDA_HOME="/path/to/(mini)conda3" | |
| bash setup/conda.sh | |
| ``` | |
| # Training | |
| ## Data format | |
| Before training, you must preprocess the training data. Before preprocessing, the data should be a `json` file, with the following format: | |
| ```json | |
| {"text": "<instance_0_text>"} | |
| {"text": "<instance_1_text>"} | |
| ``` | |
| Note that the preprocessing script will pack observations together in vectors of a specified length, and will separate each instance (json line) by the tokenizer's EOS token. | |
| Then, run the bash scripts in this order: | |
| ```bash | |
| ./preprocess_data.sh [OPTIONS] | |
| ./convert2megatron.sh [OPTIONS] | |
| ./model_sharding.sh [OPTIONS] | |
| ./continue_pretraining.sh [OPTIONS] | |
| ``` | |
| >NOTE: each of these commands may be run with flag `--help`, which will inform the user on how to use each argument. | |
| For example, for a continued pretraining run with Llama 2 7B on datasets `d1` and `d2` and 8 GPUs, run the following: | |
| ```bash | |
| > ./preprocess_data.sh --dataset_json=<path_to_d1> --dataset_bin=<d1_output_path> --vocab_file=<path_to_hf_model>/tokenizer.model --repo=<path_to_repo> | |
| > ./preprocess_data.sh --dataset_json=<path_to_d2> --dataset_bin=<d2_output_path> --vocab_file=<path_to_hf_model>/tokenizer.model --repo=<path_to_repo> | |
| > ./convert2megatron.sh --megatron_model=<megatron_model_path> --model_path=<path_to_hf_model> --size=7 --repo=<path_to_repo> | |
| > ./model_sharding.sh --megatron_model=<megatron_model_path> --sharded_model=<sharded_model_path> --tp=8 --pp=1 --vocab_size=32000 --repo=<path_to_repo> | |
| > ./continue_pretraining.sh --data_path="1 d1 1 d2" --megatron_model=<sharded_model_path> --model_dir=<checkpoint_save_dir> --tokenizer_path=<path_to_hf_model>/tokenizer.model --tp=8 --pp=1 [TRAINING_ARGS] | |
| ``` |