mirror of
https://github.com/hanzoai/open-instruct.git
synced 2026-10-30 18:26:32 +00:00
Lj MirandaandGitHub
a5fe119f7d
* Add .vscode to .gitignore * Add dependencies for synth pref pipeline * Initialize directory * Port majority of the scripts to open-instruct * Add documentation * Add public birr as a submodule * Minor fixes and update README * Add instructions for creating annotation mix * Run isort on source * Update README * Remove birr submodule * Change birr from submodule to git-checkout * Add examples * Update README with better examples
74 lines
2.6 KiB
Python
74 lines
2.6 KiB
Python
import argparse
|
|
import logging
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import yaml
|
|
from dotenv import load_dotenv
|
|
from tqdm import tqdm
|
|
|
|
from scripts.synth_pref.utils.model_configs import MODELS
|
|
|
|
logging.basicConfig(
|
|
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
|
|
datefmt="%Y-%m-%d %H:%M:%S",
|
|
handlers=[logging.StreamHandler(sys.stdout)],
|
|
level=logging.INFO,
|
|
)
|
|
|
|
load_dotenv()
|
|
|
|
|
|
def get_args():
|
|
# fmt: off
|
|
parser = argparse.ArgumentParser("Generate responses for a set of prompts given a model.")
|
|
parser.add_argument("-n", "--name", type=str, required=True, help="Name of the dataset for ID purposes.")
|
|
parser.add_argument("-f", "--source_file", nargs="+", required=True, help="Directory that contains the JSONL files you need to generate responses for.")
|
|
parser.add_argument("--target_dir", type=Path, required=True, help="Directory to save the output inference results.")
|
|
parser.add_argument("--batch_size", default=128, type=int, help="Set the number of prompts to send to vLLM at a time.")
|
|
parser.add_argument("--config_output_dir", default="configs/", help="Path to save the configuration yaml files.")
|
|
parser.add_argument("-x", "--ignore_model", nargs="+", help="Ignore a model and don't create a config from it.")
|
|
parser.add_argument("-y", "--include_model", nargs="+", help="List of models to include.")
|
|
# fmt: on
|
|
return parser.parse_args()
|
|
|
|
|
|
def main():
|
|
args = get_args()
|
|
model_names = list(MODELS.keys())
|
|
if args.include_model:
|
|
model_names = [model for model in model_names if model in args.include_model]
|
|
if args.ignore_model:
|
|
model_names = [model for model in model_names if model not in args.ignore_model]
|
|
|
|
job_yaml_paths = []
|
|
logging.info("Creating job files...")
|
|
for model in tqdm(model_names):
|
|
model_config = MODELS.get(model)
|
|
|
|
model_id = model.replace("/", "___")
|
|
task_name = f"birr-{model_id}-{args.name}"
|
|
default_pipeline_params = {
|
|
"pipeline": {
|
|
"input_file_dir": args.source_file,
|
|
"output_file_dir": args.target_dir,
|
|
"num_workers": 1,
|
|
"generation_batch_size": args.batch_size,
|
|
}
|
|
}
|
|
|
|
job = {**model_config, **default_pipeline_params}
|
|
|
|
# Save config to file
|
|
output_dir = Path(args.config_output_dir)
|
|
output_dir.mkdir(parents=True, exist_ok=True)
|
|
output_path = output_dir / f"{task_name}.yaml"
|
|
with open(output_path, "w") as file:
|
|
yaml.dump(job, file)
|
|
|
|
job_yaml_paths.append(output_path)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|