File size: 11,296 Bytes
a608d21 2710647 a608d21 2710647 a608d21 2710647 7a61837 2710647 7a61837 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 7a61837 2710647 7a61837 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 a608d21 2710647 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 | from __future__ import annotations
import argparse
import json
import sys
from pathlib import Path
from . import __version__
from .api import cache_embeddings, evaluate, run, train
from .assets import default_assets_root, discover_cached_embeddings, discover_checkpoints, discover_repo_checkpoints
from .hub import DEFAULT_REPO_ID, download_checkpoint
def _add_dataset_args(parser: argparse.ArgumentParser) -> None:
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument("--dataset", help="BEIR dataset name, such as scifact.")
group.add_argument("--dataset-path", type=Path, help="Local BEIR-format dataset directory.")
parser.add_argument("--datasets-dir", type=Path, help="BEIR download root (default: <assets>/datasets).")
def _add_encoder_args(parser: argparse.ArgumentParser) -> None:
parser.add_argument("--encoder", help="Built-in encoder alias: minilm, e5, or mpnet.")
parser.add_argument("--encoder-model-name", help="Custom SentenceTransformers model name.")
parser.add_argument("--embedding-dim", type=int, help="Required with --encoder-model-name.")
parser.add_argument("--query-prefix", default="")
parser.add_argument("--passage-prefix", default="")
parser.add_argument("--encoder-revision", help="Optional immutable revision for a custom base encoder.")
def _add_runtime_args(parser: argparse.ArgumentParser) -> None:
parser.add_argument("--device", default="cpu")
parser.add_argument("--assets-root", type=Path, default=default_assets_root())
parser.add_argument("--cache-dir", type=Path)
parser.add_argument("--max-queries", type=int)
parser.add_argument("--seed", type=int, default=42)
def _dataset_values(args: argparse.Namespace) -> tuple[str, Path]:
if args.dataset_path:
path = args.dataset_path.resolve()
return path.name, path.parent
return args.dataset, args.datasets_dir or (args.assets_root / "datasets")
def _encoder_label(args: argparse.Namespace) -> str:
return args.encoder or args.encoder_model_name or "custom"
def _cache_dir(args: argparse.Namespace, dataset: str) -> Path:
return args.cache_dir or args.assets_root / f"cache_{_encoder_label(args).replace('/', '-')}_{dataset}"
def _command_info(args: argparse.Namespace) -> int:
print(
json.dumps(
{
"version": __version__,
"default_hugging_face_repo": DEFAULT_REPO_ID,
"available_checkpoint": "minilm/scifact",
"training_recipe": {
"objective": "improvement-margin",
"improvement_margin": 0.05,
"hard_negatives": 63,
"hard_negative_method": "dense",
"hard_negative_pool": 100,
"projector_initialization": "identity",
"optimizer": "adam",
"learning_rate": 1e-4,
"weight_decay": 1e-5,
"batch_size": 32,
"epochs": 30,
"patience": 7,
"seed": 42,
},
"supported_encoders": ["minilm", "e5", "mpnet"],
},
indent=2,
)
)
return 0
def _command_list_assets(args: argparse.Namespace) -> int:
print(
json.dumps(
{
"assets_root": str(args.assets_root),
"cached_embeddings": [
item.__dict__ | {"path": str(item.path)} for item in discover_cached_embeddings(args.assets_root)
],
"workspace_checkpoints": [
item.__dict__ | {"path": str(item.path)} for item in discover_checkpoints(args.assets_root)
],
"repo_checkpoints": [
item.__dict__ | {"path": str(item.path)}
for item in discover_repo_checkpoints(Path(__file__).resolve().parents[2])
],
},
indent=2,
default=str,
)
)
return 0
def _command_download(args: argparse.Namespace) -> int:
downloaded = download_checkpoint(
encoder=args.encoder,
dataset=args.dataset,
repo_id=args.repo_id,
revision=args.revision,
cache_dir=args.hub_cache_dir,
)
print(json.dumps({"checkpoint": str(downloaded.checkpoint_path), "repo_id": downloaded.repo_id}, indent=2))
return 0
def _command_cache(args: argparse.Namespace) -> int:
dataset, datasets_dir = _dataset_values(args)
built = cache_embeddings(
encoder=args.encoder,
dataset=dataset,
cache_dir=_cache_dir(args, dataset),
datasets_dir=datasets_dir,
device=args.device,
encoder_model_name=args.encoder_model_name,
embedding_dim=args.embedding_dim,
query_prefix=args.query_prefix,
passage_prefix=args.passage_prefix,
encoder_revision=args.encoder_revision,
batch_size=args.batch_size,
num_negatives=args.num_negatives,
negative_pool=args.negative_pool,
max_queries=args.max_queries,
seed=args.seed,
)
print(json.dumps({"cache_dir": str(built), "dataset": dataset}, indent=2))
return 0
def _training_kwargs(args: argparse.Namespace) -> dict:
dataset, datasets_dir = _dataset_values(args)
return {
"encoder": args.encoder,
"dataset": dataset,
"cache_dir": _cache_dir(args, dataset),
"datasets_dir": datasets_dir,
"output_dir": args.output_dir or args.assets_root,
"device": args.device,
"encoder_model_name": args.encoder_model_name,
"embedding_dim": args.embedding_dim,
"query_prefix": args.query_prefix,
"passage_prefix": args.passage_prefix,
"encoder_revision": args.encoder_revision,
"max_queries": args.max_queries,
"batch_size": args.batch_size,
"epochs": args.epochs,
"lr": args.lr,
"weight_decay": args.weight_decay,
"num_negatives": args.num_negatives,
"feedback_k": args.candidate_k,
"ranking_k": args.ranking_k,
"k": args.k,
"improvement_margin": args.improvement_margin,
"patience": args.patience,
"seed": args.seed,
}
def _print_result(payload: dict) -> int:
print(json.dumps(payload, indent=2, default=str))
return 0
def _command_train(args: argparse.Namespace) -> int:
return _print_result(train(**_training_kwargs(args)))
def _command_run(args: argparse.Namespace) -> int:
kwargs = _training_kwargs(args)
kwargs["negative_pool"] = args.negative_pool
return _print_result(run(**kwargs))
def _command_evaluate(args: argparse.Namespace) -> int:
dataset, datasets_dir = _dataset_values(args)
return _print_result(
evaluate(
encoder=args.encoder,
dataset=dataset,
cache_dir=_cache_dir(args, dataset),
datasets_dir=datasets_dir,
checkpoint_path=args.checkpoint,
device=args.device,
encoder_model_name=args.encoder_model_name,
embedding_dim=args.embedding_dim,
query_prefix=args.query_prefix,
passage_prefix=args.passage_prefix,
encoder_revision=args.encoder_revision,
max_queries=args.max_queries,
seed=args.seed,
feedback_k=args.candidate_k,
ranking_k=args.ranking_k,
k=args.k,
)
)
def _configure_training_parser(parser: argparse.ArgumentParser) -> None:
_add_dataset_args(parser)
_add_encoder_args(parser)
_add_runtime_args(parser)
parser.add_argument("--output-dir", type=Path)
parser.add_argument("--batch-size", type=int, default=32)
parser.add_argument("--epochs", type=int, default=30)
parser.add_argument("--lr", type=float, default=1e-4)
parser.add_argument("--weight-decay", type=float, default=1e-5)
parser.add_argument("--num-negatives", type=int, default=63)
parser.add_argument("--candidate-k", type=int, default=100)
parser.add_argument("--ranking-k", type=int, default=10)
parser.add_argument("--k", type=int, default=10)
parser.add_argument("--improvement-margin", type=float, default=0.05)
parser.add_argument("--patience", type=int, default=7)
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(
prog="imrnns",
description="Train, evaluate, and use IMRNN dense-retrieval adapters.",
)
parser.add_argument("--version", action="version", version=f"%(prog)s {__version__}")
subparsers = parser.add_subparsers(dest="command", required=True)
info = subparsers.add_parser("info", help="Show package and recipe information.")
info.set_defaults(func=_command_info)
assets = subparsers.add_parser("list-assets", help="List local caches and checkpoints.")
assets.add_argument("--assets-root", type=Path, default=default_assets_root())
assets.set_defaults(func=_command_list_assets)
download = subparsers.add_parser("download", help="Download the validated adapter checkpoint.")
download.add_argument("--encoder", required=True, choices=["minilm"])
download.add_argument("--dataset", required=True, choices=["scifact"])
download.add_argument("--repo-id", default=DEFAULT_REPO_ID)
download.add_argument("--revision")
download.add_argument("--hub-cache-dir", type=Path)
download.set_defaults(func=_command_download)
cache = subparsers.add_parser("cache", help="Encode a dataset and mine dense hard negatives.")
_add_dataset_args(cache)
_add_encoder_args(cache)
_add_runtime_args(cache)
cache.add_argument("--batch-size", type=int, default=64)
cache.add_argument("--num-negatives", type=int, default=63)
cache.add_argument("--negative-pool", type=int, default=100)
cache.set_defaults(func=_command_cache)
train_parser = subparsers.add_parser("train", help="Train an adapter from a prepared cache.")
_configure_training_parser(train_parser)
train_parser.set_defaults(func=_command_train)
evaluate_parser = subparsers.add_parser("evaluate", help="Evaluate a checkpoint against its base retriever.")
_add_dataset_args(evaluate_parser)
_add_encoder_args(evaluate_parser)
_add_runtime_args(evaluate_parser)
evaluate_parser.add_argument("--checkpoint", type=Path, required=True)
evaluate_parser.add_argument("--candidate-k", type=int, default=100)
evaluate_parser.add_argument("--ranking-k", type=int, default=10)
evaluate_parser.add_argument("--k", type=int, default=10)
evaluate_parser.set_defaults(func=_command_evaluate)
run_parser = subparsers.add_parser("run", help="Cache, train, and evaluate end to end.")
_configure_training_parser(run_parser)
run_parser.add_argument("--negative-pool", type=int, default=100)
run_parser.set_defaults(func=_command_run)
return parser
def main(argv: list[str] | None = None) -> int:
parser = build_parser()
args = parser.parse_args(argv)
try:
return int(args.func(args))
except (FileNotFoundError, ImportError, RuntimeError, ValueError) as exc:
print(f"imrnns: error: {exc}", file=sys.stderr)
return 2
|