Dateien nach "app/core" hochladen
All checks were successful
Deploy mindnet to llm-node / deploy (push) Successful in 2s
All checks were successful
Deploy mindnet to llm-node / deploy (push) Successful in 2s
This commit is contained in:
parent
2a1c62aeed
commit
9e8b433c95
|
|
@ -3,14 +3,17 @@
|
|||
"""
|
||||
app/core/chunk_payload.py — Mindnet V2 (compat)
|
||||
|
||||
Ziele:
|
||||
- Bewahrt bestehendes Verhalten (index, chunk_profile, retriever_weight, etc.)
|
||||
- Denormalisiert optional `tags` aus der Note‑FM auf Chunks
|
||||
- Fügt Aliase für die Chunk‑Nummer hinzu: `ord` (v2‑Schema), `chunk_num`, `Chunk_Nummer`
|
||||
- **Kompatibilität:** akzeptiert sowohl `path_arg` (positional) als auch `file_path` (keyword)
|
||||
Ziele (ohne Bruch zur lauffähigen v1-Basis):
|
||||
- Akzeptiert `file_path=` (Alias zu path_arg)
|
||||
- Verarbeitet Chunks sowohl als `dict` **als auch** als Objekt (z. B. Dataclass `Chunk`)
|
||||
- Schreibt v1-kompatible Felder:
|
||||
* `id` (Alias von `chunk_id` – **wichtig** für app/core/edges.py v1)
|
||||
* `neighbors: {prev, next}` wird **berechnet** (Sequenz), falls nicht vorhanden
|
||||
- Denormalisiert optional `tags` der Note auf Chunks
|
||||
- Fügt Nummern-Aliase hinzu: `ord`, `chunk_num`, `Chunk_Nummer`
|
||||
|
||||
Hinweis:
|
||||
- `edge_defaults` sind Note‑Regeln (Typ) und werden nicht pro Chunk gespiegelt.
|
||||
Wichtig:
|
||||
- `edge_defaults` gehören zur *Note* (Typ-Regeln), nicht pro Chunk. Werden hier **nicht** gespiegelt.
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
|
|
@ -22,12 +25,18 @@ from typing import Any, Dict, List, Optional
|
|||
|
||||
from app.core.chunker import assemble_chunks
|
||||
|
||||
# ---------- Helpers ----------
|
||||
|
||||
def _as_dict(obj):
|
||||
if isinstance(obj, dict): return obj
|
||||
try:
|
||||
return dict(obj) # type: ignore
|
||||
except Exception:
|
||||
return {"value": obj}
|
||||
if isinstance(obj, dict):
|
||||
return obj
|
||||
# Objekt → (teilweise) Dict-Ansicht via Attribute
|
||||
d = {}
|
||||
for k in ("index","ord","chunk_index","text","window","id","chunk_id","neighbors","note_id","type","title"):
|
||||
if hasattr(obj, k):
|
||||
d[k] = getattr(obj, k)
|
||||
# Fallback: bestehe nicht auf Vollständigkeit
|
||||
return d
|
||||
|
||||
def _coalesce(*vals):
|
||||
for v in vals:
|
||||
|
|
@ -47,21 +56,21 @@ def _ensure_list(x) -> list:
|
|||
if isinstance(x, (set, tuple)): return [str(i) for i in x]
|
||||
return [str(x)]
|
||||
|
||||
def _load_types_config(types_cfg_explicit: Optional[dict] = None) -> dict:
|
||||
"""Types-Registry *optional* einspeisen (bereits geparst), sonst lazy-laden vermeiden."""
|
||||
return types_cfg_explicit or {}
|
||||
|
||||
def _text_from_note(note: Dict[str, Any]) -> str:
|
||||
# Erwartete Inputs (siehe parser.py / import_markdown.py):
|
||||
# note["body"] oder note["text"]; Fallback leerer String
|
||||
return note.get("body") or note.get("text") or ""
|
||||
|
||||
def _iter_chunks(note: Dict[str, Any], chunk_profile: str, fulltext: str) -> List[Dict[str, Any]]:
|
||||
"""Nutze bestehenden assemble_chunks(note_id, body, type)."""
|
||||
"""Nutze bestehenden assemble_chunks(note_id, body, type). Rückgabe kann Objektliste sein."""
|
||||
note_id = note.get("id") or (note.get("frontmatter") or {}).get("id")
|
||||
ntype = (note.get("frontmatter") or {}).get("type") or note.get("type") or "note"
|
||||
# assemble_chunks liefert Liste von Dicts mit mindestens {"index","text"} (v1)
|
||||
return assemble_chunks(note_id, fulltext, ntype)
|
||||
raw_list = assemble_chunks(note_id, fulltext, ntype)
|
||||
# Normalisiere auf Dicts (unter Bewahrung vorhandener Keys)
|
||||
out: List[Dict[str, Any]] = []
|
||||
for c in raw_list:
|
||||
out.append(_as_dict(c) if not isinstance(c, dict) else c)
|
||||
return out
|
||||
|
||||
# ---------- Main ----------
|
||||
|
||||
def make_chunk_payloads(
|
||||
note: Any,
|
||||
|
|
@ -73,23 +82,16 @@ def make_chunk_payloads(
|
|||
types_cfg: Optional[dict] = None,
|
||||
) -> List[Dict[str, Any]]:
|
||||
"""
|
||||
Erzeugt Chunk-Payloads. Erwartet:
|
||||
- `note`: Normalisierte Note-Struktur (inkl. frontmatter)
|
||||
- `path_arg` oder `file_path`: Pfad der Note
|
||||
- `chunks_from_chunker`: optional: Ergebnis von assemble_chunks (sonst wird intern erzeugt)
|
||||
|
||||
Rückgabe: Liste aus Payload-Dicts, jedes mit mind.:
|
||||
- note_id, chunk_id, index, ord (Alias), title, type, path, text, retriever_weight, chunk_profile
|
||||
- optional: tags (aus Note-FM), chunk_num, Chunk_Nummer (Aliases von index/ord)
|
||||
Erzeugt Chunk-Payloads im v1-kompatiblen Format (plus V2-Aliase).
|
||||
"""
|
||||
n = _as_dict(note)
|
||||
# ---- Note-Kontext ----
|
||||
n = note if isinstance(note, dict) else {"frontmatter": {}}
|
||||
fm = n.get("frontmatter") or {}
|
||||
note_type = str(fm.get("type") or n.get("type") or "note")
|
||||
types_cfg = _load_types_config(types_cfg)
|
||||
|
||||
types_cfg = types_cfg or {}
|
||||
cfg_for_type = types_cfg.get(note_type, {}) if isinstance(types_cfg, dict) else {}
|
||||
default_rw = _env_float("MINDNET_DEFAULT_RETRIEVER_WEIGHT", 1.0)
|
||||
|
||||
default_rw = _env_float("MINDNET_DEFAULT_RETRIEVER_WEIGHT", 1.0)
|
||||
retriever_weight = _coalesce(fm.get("retriever_weight"), cfg_for_type.get("retriever_weight"), default_rw)
|
||||
try:
|
||||
retriever_weight = float(retriever_weight)
|
||||
|
|
@ -101,52 +103,80 @@ def make_chunk_payloads(
|
|||
|
||||
note_id = n.get("note_id") or n.get("id") or fm.get("id")
|
||||
title = n.get("title") or fm.get("title") or ""
|
||||
# Pfad-Auflösung: Priorität file_path > note['path'] > path_arg
|
||||
|
||||
# Pfadauflösung: file_path > note['path'] > path_arg
|
||||
path = file_path or n.get("path") or path_arg
|
||||
if isinstance(path, pathlib.Path):
|
||||
path = str(path)
|
||||
path = path or "" # garantiert vorhanden
|
||||
path = path or ""
|
||||
|
||||
# Denormalisierte Tags (optional): auf Chunks spiegeln, wenn vorhanden
|
||||
# Tags denormalisieren (optional)
|
||||
tags = fm.get("tags") or fm.get("keywords") or n.get("tags")
|
||||
tags_list = _ensure_list(tags) if tags else []
|
||||
|
||||
# Quelltext
|
||||
# ---- Chunks besorgen ----
|
||||
fulltext = note_text if isinstance(note_text, str) else _text_from_note(n)
|
||||
|
||||
# Chunks besorgen
|
||||
chunks = chunks_from_chunker if isinstance(chunks_from_chunker, list) else _iter_chunks(n, chunk_profile, fulltext)
|
||||
raw_chunks = chunks_from_chunker if isinstance(chunks_from_chunker, list) else _iter_chunks(n, chunk_profile, fulltext)
|
||||
|
||||
payloads: List[Dict[str, Any]] = []
|
||||
for c in chunks:
|
||||
idx = c.get("index", len(payloads))
|
||||
text = c.get("text") if isinstance(c, dict) else (str(c) if c is not None else "")
|
||||
text = text if isinstance(text, str) else str(text or "")
|
||||
for c in raw_chunks:
|
||||
cdict = c if isinstance(c, dict) else _as_dict(c)
|
||||
# Index/Basisdaten robust ermitteln
|
||||
idx = _coalesce(cdict.get("index"), cdict.get("ord"), cdict.get("chunk_index"), len(payloads))
|
||||
try:
|
||||
idx = int(idx)
|
||||
except Exception:
|
||||
idx = len(payloads)
|
||||
|
||||
# deterministische ID (unter Beibehaltung deines bisherigen Schemas)
|
||||
text = _coalesce(cdict.get("window"), cdict.get("text"), "")
|
||||
if not isinstance(text, str):
|
||||
text = str(text or "")
|
||||
|
||||
# deterministische ID (kompatibel & stabil)
|
||||
key = f"{note_id}|{idx}"
|
||||
h = hashlib.sha1(key.encode("utf-8")).hexdigest()[:12] if note_id else hashlib.sha1(f"{path}|{idx}".encode("utf-8")).hexdigest()[:12]
|
||||
chunk_id = f"{note_id}-{idx:03d}-{h}" if note_id else f"{h}"
|
||||
chunk_id = cdict.get("chunk_id") or cdict.get("id") or (f"{note_id}-{idx:03d}-{h}" if note_id else h)
|
||||
|
||||
payload = {
|
||||
"note_id": note_id,
|
||||
"chunk_id": chunk_id,
|
||||
# v1 Kernfelder (+Erweiterungen)
|
||||
"id": chunk_id, # <— WICHTIG: v1 edges.py erwartet 'id'
|
||||
"chunk_id": chunk_id, # v2-Alias
|
||||
"index": idx,
|
||||
"ord": idx, # Alias für v2‑Schema
|
||||
"chunk_num": idx, # neutraler Alias
|
||||
"Chunk_Nummer": idx, # deutschsprachiger Alias
|
||||
"title": title,
|
||||
"ord": idx, # v2-Alias
|
||||
"chunk_num": idx,
|
||||
"Chunk_Nummer": idx,
|
||||
"note_id": note_id,
|
||||
"type": note_type,
|
||||
"path": path, # garantiert vorhanden
|
||||
"text": text, # nie leer
|
||||
"title": title,
|
||||
"path": path,
|
||||
"text": text,
|
||||
"window": text, # falls der Chunker bereits ein Fenster liefert, bleibt es identisch
|
||||
"retriever_weight": retriever_weight,
|
||||
"chunk_profile": chunk_profile,
|
||||
}
|
||||
|
||||
# Bestehende neighbors vom Chunk übernehmen (falls vorhanden)
|
||||
nb = cdict.get("neighbors")
|
||||
if isinstance(nb, dict):
|
||||
prev_id = nb.get("prev"); next_id = nb.get("next")
|
||||
payload["neighbors"] = {"prev": prev_id, "next": next_id}
|
||||
# Tags spiegeln
|
||||
if tags_list:
|
||||
payload["tags"] = tags_list
|
||||
|
||||
# JSON‑Roundtrip als einfache Validierung
|
||||
# JSON-Roundtrip als Validierung
|
||||
json.loads(json.dumps(payload, ensure_ascii=False))
|
||||
payloads.append(payload)
|
||||
|
||||
# Nachgelagert: neighbors berechnen, falls fehlend
|
||||
for i, p in enumerate(payloads):
|
||||
nb = p.get("neighbors") or {}
|
||||
prev_id = nb.get("prev")
|
||||
next_id = nb.get("next")
|
||||
if prev_id is None and i > 0:
|
||||
prev_id = payloads[i-1]["id"]
|
||||
if next_id is None and i+1 < len(payloads):
|
||||
next_id = payloads[i+1]["id"]
|
||||
p["neighbors"] = {"prev": prev_id, "next": next_id}
|
||||
|
||||
return payloads
|
||||
|
|
|
|||
Loading…
Reference in New Issue
Block a user