<img width="2318" height="1448" alt="image" src="https://github.com/user-attachments/assets/bca71d69-4972-420c-b14e-c8cf74a17de9" /> ## Summary Everything a knowledge base could not read, and the pieces it cut them into saying what they are. **Reading a PDF.** A PDF records glyphs at coordinates; whether they are a heading, a table cell or a running footer is inferred from how they look. There are now two readers. The geometric one infers structure from type size and position and needs nothing but the file. The other asks deepdoc's models, ported into `raven/knowledge/parser/deepdoc`: a layout model that says what each region is, a detector and recognizer that read a page carrying no text at all, and a table-structure model that rebuilds a grid from a picture of one. Detecting boxes around text the file already spells out exactly costs an inference pass per page to learn less, and recognising it costs accuracy. The layout graph answers with a shape its own postprocessing raises on, and its classes are DocLayout-YOLO's order rather than the list Both were established by running the model, and the label order is pinned by a test. **Tables go in as tables.** `<table>` is the only shape that keeps what a grid states -- which column a cell stood under, which cells are merged, which row is the header -- and it is the shape a model reads a table in, which is what these chunks are for. PDF, Word and PowerPoint all state their merges exactly, and they now survive as `colspan` and `rowspan`. The flat form this replaces had to repeat a header standing over two columns across both of them, because there was nowhere to record a span; it had no callers left afterwards and is gone. Two deviations from upstream, both about what the string is for: every cell's text is escaped, where upstream interpolates it raw and a cell holding a `<` produces markup that is no longer a table, and the span attributes are written only where there is a span, where upstream emits `<td >` for every ordinary cell. **A picture of where each chunk came from.** A chunk is a few hundred words with a page number on it, which is enough to cite and not enough to answer whether the parser read the page correctly. So every PDF chunk that knows where it sits carries one image: the regions it covers, cropped out of the rendered pages and stacked. here they are already structured in the parser's element spans, so the markers would only put text nobody wrote into the thing being embedded. The panel draws a chunk's tables as tables, parsed and rebuilt as elements rather than injected: a person can rewrite any piece from that same panel, so what comes back from the store is user input however it got there. **The page that reads them.** Upstream retired the knowledge page while this branch was out; it is rebuilt against the new architecture, and the half that never existed is the one a reader needs -- what a document actually became once it was cut and embedded. Documents sit in one-level folders and arrive by four routes: a file, a folder, a URL, a written note. Adding and indexing are two calls on the contract and nothing drains that queue on its own, so a page that stopped after the first left every upload saying Queued for good. Opening a file takes the whole page: the document on the left, the pieces the index holds of it on the right. The left half is the file drawn as its own pages rather than framed by the browser's PDF viewer, which reads where to open once and answers to nothing afterwards -- and which nothing can be drawn on. Pages are rendered once and kept, and the region a piece was cut from is washed over the page it came from, placed in fractions of the page so the picture's own scale never enters the arithmetic. Clicking a piece takes the file there; double-clicking opens it for rewriting. The pieces can be searched, filtered by state, turned on and off, deleted, rewritten and added to, one at a time or in a batch. Every one of those was already on the contract with no reader for it. A base's settings are back and fuller: how many pieces a search returns, how documents are cut, and how much prose comes with a table or a figure. The embedding model is a picker while the base holds no pieces and a stated fact once it does -- moving it then is the collection made again at the new width and every document re-cut, which is not a thing a panel does by being saved. **An agent can read them.** The retrieval engine was there and nothing could reach it: no agent had a tool that opened a library. A reader now points a conversation at bases from the composer, the way they pin a working directory, and the selection rides in the same place -- the session's metadata, bound around the turn. The tool takes a query and never a base. The scope is the reader's choice, so a model cannot reach material nobody offered it, and it is withheld where nothing was picked rather than offered and refused. A hit says which document and page it came from, because a hit a reader cannot place is a hit they cannot check -- and the passages are pressable in the transcript, walking into the base, the document, the folder it is filed under and then the piece. A citation a reader cannot follow is a claim they have to take on trust, and the whole point of searching their own material is that they can go and look. **The word segmenter and the resources.** `raven.core.Tokenizer` is Infinity's `rag_tokenizer`, ported: a forward and backward maximum match over a dictionary trie, with stemming and lemmatisation on the latin path. XGBoost is not carried over. None of the data lives in git -- the dictionary, the NLTK corpora and the ONNX graphs run to about a hundred megabytes, so `make fetch-resources` on an install and an optional Docker layer behind `RAVEN_RESOURCES` both point at one script. A checkout that could not reach the network still works, and every feature that reads this data says so when asked. ## Type - [ ] Fix - [x] Feature - [ ] Docs - [ ] CI / tooling - [ ] Refactor - [ ] Other ## Verification Run on macOS against Python 3.12, with the deepdoc weights fetched (the model tests skip without them): ``` uv run pytest 27450 passed, 111 skipped, 3 failed npm run test --prefix ui-web 3276 passed npm run test --prefix ui-tui 2106 passed npm run type-check --prefix ui-web clean (tsc) npm run lint --prefix ui-web clean (4 pre-existing hook warnings) make lint-python clean (ruff check + format) make lint-types clean (ty) make lint-imports 10 contracts kept, 0 broken make check-source-language clean make check-large-files clean make build-ui built, both boot snapshots match node ui-web/scripts/check-css.mjs clean node ui-web/scripts/check-class-namespace.mjs clean ``` The three Python failures are not this branch's and were each confirmed on the base alone: one asks a host without LibreOffice to refuse a render and this host has LibreOffice installed, and two simulation cases fail on an unmodified checkout of the base commit. The RPC contract was regenerated for both consumers and both match it (`npm run gen --prefix ui-web`, `npm run gen:rpc --prefix ui-tui`, then `gen:check`). The page's boot snapshots and region goldens were refreshed where the composer bar gained a button, and each records that button and nothing else. - [x] Relevant tests pass locally - [x] Relevant lint / type checks pass locally - [x] User-facing docs or screenshots are updated when needed ## Risk A PDF's tables, and a Word or PowerPoint file's, are now indexed as HTML rather than as header-prefixed lines. Chunk text changes shape for those documents, so a base indexed before this and searched after it holds both forms until its documents are reindexed; retrieval keeps working either way, because both forms are text. The same applies to how much prose a table or a figure carries with it: the setting now reaches as far as it says, where before it stopped at the nearest paragraph, so a chunk cut before this keeps the narrower context until its document is rebuilt. Rolling back is reverting the branch and reindexing. The deepdoc models are optional. An install that never ran `make fetch-resources` reads PDFs geometrically, exactly as before, and says so when the models are asked for by name. Nothing here reaches the network at request time. Chunk crops and rendered pages are written at index time and on first view, both under the state directory rather than in the repository, and both are deleted with their document. A page is immutable for the life of its document id, so the route serving them caches; the document route does not. All three routes take ids and no path, and the two image routes answer under the strictest sandbox in the tree. What an agent may search is the reader's choice and nothing else. The tool takes no base argument, so a model cannot name one, and it is absent from the schema entirely for a conversation that named none -- a lane nobody pointed at a base is unchanged by this branch. The bases a session carries are stored beside its working directory and travel no further. - [x] Security impact considered - [x] Backward compatibility considered - [x] Rollback path is clear for risky changes ## Related Issues N/A --------- Co-authored-by: Claude (claude-opus-5) <noreply@anthropic.com>
183 lines
7.3 KiB
Python
183 lines
7.3 KiB
Python
"""The table-structure model: what it finds that a line finder cannot.
|
|
|
|
Skipped whole when the weights are not installed, which is a checkout where
|
|
`make fetch-resources` has not run.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
|
|
import pytest
|
|
|
|
from raven.knowledge.parser import ELEMENTS, LAYOUT_TYPE, LayoutType
|
|
from raven.knowledge.parser._pdfdoc import open_document
|
|
from raven.knowledge.parser.deepdoc import _onnx, _pipeline
|
|
from raven.knowledge.parser.pdf_parser import PdfParser
|
|
|
|
pytestmark = pytest.mark.skipif(
|
|
not _onnx.available("layout", "tsr"), reason="the deepdoc layout and table models are not installed"
|
|
)
|
|
|
|
_GRID = [
|
|
["Region", "Q1", "Q2"],
|
|
["EU", "1.2M", "1.4M"],
|
|
["US", "2.4M", "2.6M"],
|
|
["APAC", "0.8M", "1.1M"],
|
|
]
|
|
|
|
|
|
def _bytes(build) -> bytes:
|
|
"""A one-page PDF built by the fixture, as bytes."""
|
|
import pymupdf
|
|
|
|
document = pymupdf.open()
|
|
build(document.new_page(), pymupdf)
|
|
raw = document.tobytes()
|
|
document.close()
|
|
return raw
|
|
|
|
|
|
def _ruled(page, pymupdf) -> None:
|
|
"""A table drawn with its lines, the kind a line finder can see."""
|
|
page.insert_text(pymupdf.Point(60, 70), "Revenue by region", fontsize=18)
|
|
xs, ys = [60, 200, 340, 480], [110, 145, 180, 215, 250]
|
|
for y in ys:
|
|
page.draw_line(pymupdf.Point(xs[0], y), pymupdf.Point(xs[-1], y))
|
|
for x in xs:
|
|
page.draw_line(pymupdf.Point(x, ys[0]), pymupdf.Point(x, ys[-1]))
|
|
for r, row in enumerate(_GRID):
|
|
for c, value in enumerate(row):
|
|
page.insert_text(pymupdf.Point(xs[c] + 8, ys[r] + 23), value, fontsize=11)
|
|
|
|
|
|
def _unruled(page, pymupdf) -> None:
|
|
"""The same table set with whitespace, the way a paper sets one."""
|
|
page.insert_text(pymupdf.Point(60, 70), "Revenue by region", fontsize=18)
|
|
for r, row in enumerate(_GRID):
|
|
for c, value in enumerate(row):
|
|
page.insert_text(pymupdf.Point(60 + c * 140, 130 + r * 35), value, fontsize=11)
|
|
|
|
|
|
# -- what the model is asked ---------------------------------------
|
|
|
|
|
|
def test_the_model_finds_the_rows_and_columns_of_an_unruled_table() -> None:
|
|
"""The whole argument for the weights. The line-based finder looks for
|
|
ruling lines, so a table set with whitespace is invisible to it; the model
|
|
looks at the table the way a reader does."""
|
|
document = open_document(_bytes(_unruled))
|
|
page = document[0]
|
|
|
|
assert not page.tables(), "no lines to find, which is the point"
|
|
|
|
structure = _pipeline._table_model()([_pipeline.render(page)])[0]
|
|
labels = [box["label"] for box in structure]
|
|
|
|
assert labels.count("table row") >= len(_GRID)
|
|
assert labels.count("table column") >= len(_GRID[0])
|
|
|
|
|
|
# -- what it buys --------------------------------------------------
|
|
|
|
|
|
def test_a_table_is_rebuilt_as_a_grid() -> None:
|
|
"""The model says where the rows and columns are; the text layer says what
|
|
the cells are. Put back together as a table rather than flattened, because
|
|
a grid is what the thing was and what a model reads it as."""
|
|
sections = asyncio.run(PdfParser().parse(_bytes(_ruled), "revenue.pdf"))
|
|
|
|
text = " ".join(section.content.text for section in sections)
|
|
# The column names as a header row: the reader that follows the ruled lines
|
|
# has always said so, and this one says it too.
|
|
assert "<tr><th>Region</th><th>Q1</th><th>Q2</th></tr>" in text
|
|
assert "<tr><td>EU</td><td>1.2M</td><td>1.4M</td></tr>" in text
|
|
kinds = {span[LAYOUT_TYPE] for section in sections for span in section.metadata[ELEMENTS]}
|
|
assert LayoutType.TABLE in kinds
|
|
|
|
|
|
def test_a_cell_carrying_markup_cannot_break_out_of_its_cell() -> None:
|
|
"""The text comes off a page, and the table it lands in is indexed, handed
|
|
to a model and rendered in a panel. Upstream interpolates it raw."""
|
|
|
|
def build(page, pymupdf):
|
|
page.insert_text(pymupdf.Point(60, 70), "Revenue by region", fontsize=18)
|
|
xs, ys = [60, 200, 340, 480], [110, 145, 180, 215]
|
|
for y in ys:
|
|
page.draw_line(pymupdf.Point(xs[0], y), pymupdf.Point(xs[-1], y))
|
|
for x in xs:
|
|
page.draw_line(pymupdf.Point(x, ys[0]), pymupdf.Point(x, ys[-1]))
|
|
rows = [["Region", "<b>Q1</b>", "Q2"], ["EU & co", "1.2M", "1.4M"], ["US", "2.4M", "2.6M"]]
|
|
for r, row in enumerate(rows):
|
|
for c, value in enumerate(row):
|
|
page.insert_text(pymupdf.Point(xs[c] + 8, ys[r] + 23), value, fontsize=11)
|
|
|
|
text = " ".join(s.content.text for s in asyncio.run(PdfParser().parse(_bytes(build), "r.pdf")))
|
|
|
|
assert "<b>Q1</b>" in text
|
|
assert "<b>" not in text
|
|
assert "EU & co" in text
|
|
|
|
|
|
def test_a_cell_is_not_also_indexed_as_a_line_of_prose() -> None:
|
|
"""A table's cells are text boxes too. Indexed twice, a search for a figure
|
|
matches both the row that explains it and a bare number that does not."""
|
|
sections = asyncio.run(PdfParser().parse(_bytes(_ruled), "revenue.pdf"))
|
|
|
|
text = " ".join(section.content.text for section in sections)
|
|
assert text.count("2.4M") == 1
|
|
|
|
|
|
# -- when it is not there ------------------------------------------
|
|
|
|
|
|
def test_a_table_the_model_chokes_on_is_read_as_prose(monkeypatch) -> None:
|
|
"""Worse than a grid, better than nothing, and not worth the document."""
|
|
|
|
def explode():
|
|
raise RuntimeError("the graph refused this crop")
|
|
|
|
monkeypatch.setattr(_pipeline, "_table_model", explode)
|
|
|
|
sections = asyncio.run(PdfParser().parse(_bytes(_ruled), "revenue.pdf"))
|
|
|
|
text = " ".join(section.content.text for section in sections)
|
|
assert "2.4M" in text, "the cells were still read"
|
|
assert "Region: EU" not in text, "just not as a grid"
|
|
|
|
|
|
def test_an_absent_model_leaves_the_rest_of_the_layout_working(monkeypatch) -> None:
|
|
"""The three models are fetched together, but one of them missing is not a
|
|
reason to stop asking the other two."""
|
|
monkeypatch.setattr(_pipeline, "available", lambda *names: "tsr" not in names and _onnx.available(*names))
|
|
|
|
sections = asyncio.run(PdfParser().parse(_bytes(_ruled), "revenue.pdf"))
|
|
|
|
assert [s.metadata.get("heading_path") for s in sections] == [["Revenue by region"]]
|
|
|
|
|
|
def test_both_readers_render_one_table_the_same_way(monkeypatch) -> None:
|
|
"""The models are optional, so the same document must not index one way on
|
|
a machine that has them and another way on a machine that does not.
|
|
|
|
The two readers find a table's grid by different means -- one asks the
|
|
structure model, the other follows the ruled lines -- and they used to
|
|
disagree about the header row: upstream's rule looks for a header REPEATED
|
|
down a long table by spotting a row that breaks the body's dominant cell
|
|
type, which finds nothing at all when the body is words, so the column
|
|
names came back as body cells. A base built on one machine then read
|
|
differently on the other.
|
|
"""
|
|
raw = _bytes(_ruled)
|
|
|
|
with_models = asyncio.run(PdfParser().parse(raw, "revenue.pdf"))
|
|
monkeypatch.setattr(_pipeline, "available", lambda *names: False)
|
|
without = asyncio.run(PdfParser().parse(raw, "revenue.pdf"))
|
|
|
|
def table_of(sections) -> str:
|
|
text = "\n".join(section.content.text for section in sections)
|
|
return text[text.index("<table>") : text.index("</table>") + len("</table>")]
|
|
|
|
assert table_of(with_models) == table_of(without)
|
|
assert "<th>" in table_of(without), "the column names are a header row, not body"
|