Spaces:
Running on Zero
Running on Zero
File size: 8,707 Bytes
86e797b | 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 | """Small command-line entry point; paths are relative to the working directory."""
import json
from pathlib import Path
from typing import Annotated
import typer
from visual_doc_rag import data
app = typer.Typer(no_args_is_help=True, pretty_exceptions_enable=False)
Subset = Annotated[str, typer.Option(help="Dataset subset: hr or industrial.")]
DataDir = Annotated[Path, typer.Option(help="Local data/cache directory; ignored by Git.")]
def check_subset(subset: str) -> None:
if subset not in data.SUBSETS:
raise typer.BadParameter(f"Choose one of: {', '.join(data.SUBSETS)}", param_hint="--subset")
@app.command()
def build_demo(
subset: Subset = "hr",
data_dir: DataDir = Path("data"),
output: Annotated[Path | None, typer.Option(help="New bundle directory.")] = None,
model: Annotated[
str, typer.Option(help="Model used by the existing sample cache.")
] = "gemini-3.5-flash-lite",
space: Annotated[bool, typer.Option(help="Include Hugging Face Space configuration.")] = False,
) -> None:
"""Package offline retrieval and five cached samples, without calling the provider."""
from visual_doc_rag.demo import build_demo as pack
check_subset(subset)
try:
result = pack(subset, data_dir, output, model, space)
except Exception as error:
typer.echo(f"Demo build failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(f"Demo bundle: {result}")
@app.command()
def download(
subset: Subset = "hr",
data_dir: DataDir = Path("data"),
revision: Annotated[
str, typer.Option(help="HF branch, tag or commit; resolved to a SHA.")
] = "main",
) -> None:
"""Download corpus, queries, qrels and document metadata (no original PDFs)."""
check_subset(subset)
try:
manifest = data.download(subset, data_dir, revision)
except Exception as error:
typer.echo(f"Download failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(json.dumps(manifest, indent=2))
@app.command()
def inspect(subset: Subset = "hr", data_dir: DataDir = Path("data")) -> None:
"""Inspect local tables, validate joins and save aggregate statistics/sample pages."""
check_subset(subset)
try:
report = data.inspect(subset, data_dir)
except Exception as error:
typer.echo(f"Inspection failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(json.dumps({k: v for k, v in report.items() if k != "features"}, indent=2))
typer.echo(f"Saved schema and statistics: {data_dir / subset / 'inspection.json'}")
if not all(report["integrity_checks"].values()):
typer.echo("Integrity checks failed; review inspection.json before indexing.", err=True)
raise typer.Exit(1)
@app.command()
def pages(subset: Subset = "hr", data_dir: DataDir = Path("data")) -> None:
"""Export all original page images to a revision-aware PNG cache."""
from visual_doc_rag.pages import export_pages
check_subset(subset)
try:
result = export_pages(subset, data_dir)
except Exception as error:
typer.echo(f"Page export failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(f"Exported {len(result['pages'])} pages; {result['image_bytes'] / 1e6:.1f} MB")
@app.command()
def index(
subset: Subset = "hr",
data_dir: DataDir = Path("data"),
device: Annotated[str, typer.Option(help="auto, cuda or cpu for page encoding.")] = "auto",
mode: Annotated[str, typer.Option(help="all, text or visual.")] = "all",
model_revision: Annotated[
str, typer.Option(help="Adapter revision; reuse cached pin by default.")
] = "main",
base_revision: Annotated[
str | None, typer.Option(help="Optional pinned base-model revision.")
] = None,
) -> None:
"""Build BM25 and visual indexes; reuse completed page embeddings on retry."""
import torch
from visual_doc_rag.text_index import TextIndex
from visual_doc_rag.visual_index import build_visual
check_subset(subset)
if mode not in {"all", "text", "visual"} or device not in {"auto", "cuda", "cpu"}:
raise typer.BadParameter("Use --mode all/text/visual and --device auto/cuda/cpu.")
torch.set_num_threads(4)
if device == "auto":
device = "cuda" if torch.cuda.is_available() else "cpu"
try:
source, tables = data.load_tables(subset, data_dir)
if mode in {"all", "text"}:
result = TextIndex.build(tables["corpus"], data_dir / subset / "indexes", source)
typer.echo(f"BM25: {json.dumps(result)}")
if mode in {"all", "visual"}:
result = build_visual(subset, data_dir, source, device, model_revision, base_revision)
typer.echo(json.dumps({k: v for k, v in result.items() if k != "pages"}, indent=2))
except Exception as error:
typer.echo(f"Index build failed: {error}", err=True)
raise typer.Exit(1) from error
@app.command(name="eval")
def eval_command(
subset: Subset = "hr",
data_dir: DataDir = Path("data"),
runs_dir: Annotated[
Path, typer.Option(help="Directory for timestamped metric reports.")
] = Path("runs"),
) -> None:
"""Evaluate text/visual/hybrid on every English query, with CPU latency."""
import torch
from visual_doc_rag.evaluate import evaluate, format_table
check_subset(subset)
torch.set_num_threads(4)
try:
output, report = evaluate(subset, data_dir, runs_dir)
except Exception as error:
typer.echo(f"Evaluation failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(format_table(report))
typer.echo(f"Queries: {report['n_queries']}; pages: {report['n_pages']}")
typer.echo(f"Saved: {output / 'metrics.json'}")
@app.command()
def search(
query: Annotated[str, typer.Argument(help="Question or keywords to retrieve pages for.")],
subset: Subset = "hr",
data_dir: DataDir = Path("data"),
method: Annotated[str, typer.Option(help="text, visual or hybrid.")] = "hybrid",
top_k: Annotated[int, typer.Option(min=1, help="Number of page results to display.")] = 5,
export: Annotated[
bool, typer.Option(help="Save an HTML report, page images and JSON under runs/search/.")
] = False,
) -> None:
"""Retrieve pages and print local image paths, without generating an answer."""
from visual_doc_rag.retrieve import PageRetriever
check_subset(subset)
if method not in {"text", "visual", "hybrid"} or not query.strip():
raise typer.BadParameter("Provide a nonempty query and --method text/visual/hybrid.")
try:
retriever = PageRetriever(subset, data_dir)
results = retriever.search(query, method, top_k)
output = {"method": method, "pages": results}
if export:
from visual_doc_rag.search_export import export_search
output["exported_report"] = str(
export_search(query, method, retriever.source, results, retriever.tables["corpus"])
)
except Exception as error:
typer.echo(f"Search failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(json.dumps(output, indent=2))
@app.command()
def ask(
query: Annotated[str, typer.Argument(help="Question to answer using retrieved page images.")],
subset: Subset = "hr",
data_dir: DataDir = Path("data"),
method: Annotated[str, typer.Option(help="text, visual or hybrid.")] = "hybrid",
export: Annotated[bool, typer.Option(help="Save an answer and evidence HTML report.")] = False,
) -> None:
"""Answer from the top-3 pages with validated citations (requires GEMINI_API_KEY)."""
from visual_doc_rag.serve import AnswerService
check_subset(subset)
if method not in {"text", "visual", "hybrid"} or not query.strip():
raise typer.BadParameter("Provide a nonempty query and --method text/visual/hybrid.")
try:
service = AnswerService(subset, data_dir)
output = service.ask(query, method)
if export:
from visual_doc_rag.search_export import export_search
output["exported_report"] = str(
export_search(
output["query"],
method,
service.retriever.source,
output["pages"],
service.retriever.tables["corpus"],
answer=output,
)
)
except Exception as error:
typer.echo(f"Answer failed: {error}", err=True)
raise typer.Exit(1) from error
typer.echo(json.dumps(output, indent=2, ensure_ascii=False))
|