Spaces:
Running
feat(adapters/ocr): Sprint A14-S32 — MistralOCRAdapter natif (no shim)
Browse filesMigration native du legacy picarones.engines.mistral_ocr vers le contrat
BaseOCRAdapter (S26). Pas un shim : implémente directement le contrat du
nouveau monde.
picarones/adapters/ocr/mistral_ocr.py
-------------------------------------
- MistralOCRAdapter(BaseOCRAdapter) — direct hérite, pas de shim.
- input_types = {IMAGE}, output_types = {RAW_TEXT}, execution_mode = "io"
(appels HTTP, ThreadPool dans le runner).
- Constructeur kwargs-only : name, model (défaut mistral-ocr-latest),
prompt, max_tokens, api_key (override de MISTRAL_API_KEY env var),
timeout_seconds. Validation stricte au constructeur.
- Routing automatique selon model :
· mistral-ocr-* → POST /v1/ocr via urllib (endpoint dédié, retourne
pages structurées en markdown) ;
· pixtral-* → API chat/vision via SDK mistralai (texte transcrit
par message du LLM).
- _resolve_api_key() : explicite > env var > OCRAdapterError.
- _encode_image() : détection MIME selon extension (png/jpg/tif/webp,
fallback jpeg pour les inconnues), base64 + data URI.
- execute() : valide IMAGE input + URI + fichier ; route selon model ;
écrit dans <stem>.<name>.txt à côté de l'image ; retourne Artifact
RAW_TEXT.
- Erreurs HTTP/SDK wrappées dans OCRAdapterError avec type+message.
- SDK mistralai absent → OCRAdapterError clair avec pip install.
Tests S32 dédiés (28 nouveaux)
------------------------------
- Constructor : defaults, custom name/model, rejet name vide/invalide,
rejet max_tokens et timeout_seconds non-positifs.
- Contract : isinstance BaseOCRAdapter, input/output_types,
execution_mode = "io".
- ApiKey : explicite prend priorité, env utilisé sans explicite, pas de
clé → OCRAdapterError.
- Encoding : png → image/png, jpg → image/jpeg, extension inconnue →
défaut jpeg.
- InputValidation : IMAGE absent, sans URI, image inexistante, pas de
clé API → tous OCRAdapterError.
- NativeAPI : concatène pages markdown avec \n\n, écrit à
<stem>.<name>.txt, ConnectionError → OCRAdapterError wrappé.
- VisionAPI : pixtral-* route vers chat/vision SDK, SDK manquant →
OCRAdapterError, RuntimeError SDK → OCRAdapterError wrappé.
- ArtifactID : utilise adapter name dans id "<doc>:<name>:raw_text".
Pas de confidences pour S32
---------------------------
Confidences (legacy S49 cascade pages.{words,lines,blocks}.confidence)
reportées au sprint dédié ConfidenceArtifact.
Tests : 4671 passed, 11 skipped (vs 4643 avant : +28 S32).
Lint : ruff check picarones/ tests/ → All checks passed.
https://claude.ai/code/session_011XQZNitg1rCgia8ZD1a2hP
|
@@ -396,7 +396,7 @@ ruff check picarones/ tests/
|
|
| 396 |
python -m mypy picarones/core/
|
| 397 |
```
|
| 398 |
|
| 399 |
-
**Test suite**: ~
|
| 400 |
floor at 85% (currently ~87%). The `network` marker excludes tests
|
| 401 |
requiring live HTTP. A handful of tests depend on optional engines
|
| 402 |
(`pero-ocr`, `pytesseract`) and are skipped/fail gracefully when
|
|
|
|
| 396 |
python -m mypy picarones/core/
|
| 397 |
```
|
| 398 |
|
| 399 |
+
**Test suite**: ~4690 tests, ~3 min on a modern laptop. Coverage
|
| 400 |
floor at 85% (currently ~87%). The `network` marker excludes tests
|
| 401 |
requiring live HTTP. A handful of tests depend on optional engines
|
| 402 |
(`pero-ocr`, `pytesseract`) and are skipped/fail gracefully when
|
|
@@ -20,6 +20,7 @@ dédiés, **natifs** au nouveau contrat (pas de shim sur le legacy
|
|
| 20 |
from __future__ import annotations
|
| 21 |
|
| 22 |
from picarones.adapters.ocr.base import BaseOCRAdapter, OCRAdapterError
|
|
|
|
| 23 |
from picarones.adapters.ocr.pero_ocr import PeroOCRAdapter
|
| 24 |
from picarones.adapters.ocr.precomputed import PrecomputedTextAdapter
|
| 25 |
from picarones.adapters.ocr.tesseract import TesseractAdapter
|
|
@@ -27,6 +28,7 @@ from picarones.adapters.ocr.tesseract import TesseractAdapter
|
|
| 27 |
__all__ = [
|
| 28 |
"BaseOCRAdapter",
|
| 29 |
"OCRAdapterError",
|
|
|
|
| 30 |
"PeroOCRAdapter",
|
| 31 |
"PrecomputedTextAdapter",
|
| 32 |
"TesseractAdapter",
|
|
|
|
| 20 |
from __future__ import annotations
|
| 21 |
|
| 22 |
from picarones.adapters.ocr.base import BaseOCRAdapter, OCRAdapterError
|
| 23 |
+
from picarones.adapters.ocr.mistral_ocr import MistralOCRAdapter
|
| 24 |
from picarones.adapters.ocr.pero_ocr import PeroOCRAdapter
|
| 25 |
from picarones.adapters.ocr.precomputed import PrecomputedTextAdapter
|
| 26 |
from picarones.adapters.ocr.tesseract import TesseractAdapter
|
|
|
|
| 28 |
__all__ = [
|
| 29 |
"BaseOCRAdapter",
|
| 30 |
"OCRAdapterError",
|
| 31 |
+
"MistralOCRAdapter",
|
| 32 |
"PeroOCRAdapter",
|
| 33 |
"PrecomputedTextAdapter",
|
| 34 |
"TesseractAdapter",
|
|
@@ -0,0 +1,312 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""``MistralOCRAdapter`` natif — Sprint A14-S32.
|
| 2 |
+
|
| 3 |
+
Migration native du legacy ``picarones.engines.mistral_ocr.MistralOCREngine``
|
| 4 |
+
vers le contrat ``BaseOCRAdapter`` (S26). **Pas un shim** : la classe
|
| 5 |
+
implémente directement le contrat du nouveau monde.
|
| 6 |
+
|
| 7 |
+
Le legacy ``MistralOCREngine`` reste en place jusqu'au S46.
|
| 8 |
+
|
| 9 |
+
Cas d'usage BnF
|
| 10 |
+
---------------
|
| 11 |
+
Mistral AI fournit deux familles d'OCR :
|
| 12 |
+
|
| 13 |
+
- **API dédiée ``/v1/ocr``** pour les modèles ``mistral-ocr-*`` —
|
| 14 |
+
endpoint optimisé qui renvoie des pages structurées en markdown
|
| 15 |
+
(et parfois des confidences mot par mot).
|
| 16 |
+
- **API vision/chat** pour les modèles ``pixtral-*`` —
|
| 17 |
+
reconnaissance via prompt textuel + image base64.
|
| 18 |
+
|
| 19 |
+
L'adapter route automatiquement selon le nom du modèle.
|
| 20 |
+
|
| 21 |
+
Configuration
|
| 22 |
+
-------------
|
| 23 |
+
Constructeur :
|
| 24 |
+
|
| 25 |
+
- ``name`` (défaut ``"mistral_ocr"``) : identifiant de l'instance.
|
| 26 |
+
- ``model`` (défaut ``"mistral-ocr-latest"``) : modèle Mistral.
|
| 27 |
+
- ``mistral-ocr-*`` → endpoint dédié ;
|
| 28 |
+
- ``pixtral-*`` → API vision/chat.
|
| 29 |
+
- ``prompt`` : texte du prompt pour les modèles vision. Défaut :
|
| 30 |
+
instruction générique de transcription.
|
| 31 |
+
- ``max_tokens`` (défaut 4096) : limite tokens en sortie pour les
|
| 32 |
+
modèles vision.
|
| 33 |
+
- ``api_key`` : clé API Mistral. Si ``None`` (défaut), lit la
|
| 34 |
+
variable d'environnement ``MISTRAL_API_KEY``.
|
| 35 |
+
- ``timeout_seconds`` (défaut 60) : timeout HTTP pour ``urllib``.
|
| 36 |
+
|
| 37 |
+
Comportement
|
| 38 |
+
------------
|
| 39 |
+
1. Vérifie présence d'un ``Artifact`` ``IMAGE`` avec URI valide.
|
| 40 |
+
2. Encode l'image en base64 + détecte ``image/...`` MIME selon
|
| 41 |
+
l'extension.
|
| 42 |
+
3. Route vers ``/v1/ocr`` ou chat/vision selon ``model``.
|
| 43 |
+
4. Concatène le markdown / texte de toutes les pages.
|
| 44 |
+
5. Écrit dans ``<stem>.<name>.txt`` à côté de l'image.
|
| 45 |
+
6. Retourne un ``Artifact`` ``RAW_TEXT``.
|
| 46 |
+
|
| 47 |
+
Anti-sur-ingénierie
|
| 48 |
+
-------------------
|
| 49 |
+
- Pas de retry / backoff (le caller wrappe si besoin).
|
| 50 |
+
- Pas d'extraction de confidences (legacy S49 — reportées au
|
| 51 |
+
sprint ``ConfidenceArtifact``).
|
| 52 |
+
- Pas de support multi-page (l'image est traitée comme une seule
|
| 53 |
+
page d'entrée — Mistral OCR retourne une liste de pages dont on
|
| 54 |
+
concatène les markdowns).
|
| 55 |
+
"""
|
| 56 |
+
|
| 57 |
+
from __future__ import annotations
|
| 58 |
+
|
| 59 |
+
import base64
|
| 60 |
+
import json
|
| 61 |
+
import os
|
| 62 |
+
import urllib.request
|
| 63 |
+
from pathlib import Path
|
| 64 |
+
from typing import Any
|
| 65 |
+
|
| 66 |
+
from picarones.adapters.ocr.base import BaseOCRAdapter, OCRAdapterError
|
| 67 |
+
from picarones.domain.artifacts import Artifact, ArtifactType
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
_DEFAULT_PROMPT = (
|
| 71 |
+
"Transcris fidèlement le texte visible sur cette image de document "
|
| 72 |
+
"historique. Retourne uniquement le texte, sans commentaire."
|
| 73 |
+
)
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
_MEDIA_TYPES: dict[str, str] = {
|
| 77 |
+
".jpg": "image/jpeg",
|
| 78 |
+
".jpeg": "image/jpeg",
|
| 79 |
+
".png": "image/png",
|
| 80 |
+
".tif": "image/tiff",
|
| 81 |
+
".tiff": "image/tiff",
|
| 82 |
+
".webp": "image/webp",
|
| 83 |
+
}
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
class MistralOCRAdapter(BaseOCRAdapter):
|
| 87 |
+
"""Adapter Mistral OCR natif au contrat S26.
|
| 88 |
+
|
| 89 |
+
Parameters
|
| 90 |
+
----------
|
| 91 |
+
name:
|
| 92 |
+
Identifiant lisible. Défaut ``"mistral_ocr"``.
|
| 93 |
+
model:
|
| 94 |
+
Modèle Mistral. ``mistral-ocr-*`` → API dédiée ``/v1/ocr``,
|
| 95 |
+
``pixtral-*`` → API vision/chat. Défaut ``"mistral-ocr-latest"``.
|
| 96 |
+
prompt:
|
| 97 |
+
Prompt pour les modèles vision.
|
| 98 |
+
max_tokens:
|
| 99 |
+
Limite tokens en sortie pour les modèles vision. Défaut 4096.
|
| 100 |
+
api_key:
|
| 101 |
+
Clé API Mistral. Si ``None`` (défaut), lit
|
| 102 |
+
``MISTRAL_API_KEY``.
|
| 103 |
+
timeout_seconds:
|
| 104 |
+
Timeout HTTP pour les appels ``urllib``. Défaut 60.
|
| 105 |
+
|
| 106 |
+
Raises
|
| 107 |
+
------
|
| 108 |
+
OCRAdapterError
|
| 109 |
+
Si ``name`` est invalide au constructeur.
|
| 110 |
+
"""
|
| 111 |
+
|
| 112 |
+
input_types = frozenset({ArtifactType.IMAGE})
|
| 113 |
+
output_types = frozenset({ArtifactType.RAW_TEXT})
|
| 114 |
+
execution_mode = "io"
|
| 115 |
+
|
| 116 |
+
def __init__(
|
| 117 |
+
self,
|
| 118 |
+
*,
|
| 119 |
+
name: str = "mistral_ocr",
|
| 120 |
+
model: str = "mistral-ocr-latest",
|
| 121 |
+
prompt: str = _DEFAULT_PROMPT,
|
| 122 |
+
max_tokens: int = 4096,
|
| 123 |
+
api_key: str | None = None,
|
| 124 |
+
timeout_seconds: float = 60.0,
|
| 125 |
+
) -> None:
|
| 126 |
+
if not name or not name.strip():
|
| 127 |
+
raise OCRAdapterError(
|
| 128 |
+
"MistralOCRAdapter : name vide non autorisé.",
|
| 129 |
+
)
|
| 130 |
+
if not all(c.isalnum() or c in "_-" for c in name):
|
| 131 |
+
raise OCRAdapterError(
|
| 132 |
+
f"MistralOCRAdapter : name invalide {name!r} — "
|
| 133 |
+
"alphanumérique + _ - uniquement.",
|
| 134 |
+
)
|
| 135 |
+
if max_tokens <= 0:
|
| 136 |
+
raise OCRAdapterError(
|
| 137 |
+
f"MistralOCRAdapter : max_tokens doit être > 0, "
|
| 138 |
+
f"reçu {max_tokens}.",
|
| 139 |
+
)
|
| 140 |
+
if timeout_seconds <= 0:
|
| 141 |
+
raise OCRAdapterError(
|
| 142 |
+
f"MistralOCRAdapter : timeout_seconds doit être > 0, "
|
| 143 |
+
f"reçu {timeout_seconds}.",
|
| 144 |
+
)
|
| 145 |
+
self._name = name
|
| 146 |
+
self._model = model
|
| 147 |
+
self._prompt = prompt
|
| 148 |
+
self._max_tokens = max_tokens
|
| 149 |
+
self._explicit_api_key = api_key
|
| 150 |
+
self._timeout = timeout_seconds
|
| 151 |
+
|
| 152 |
+
@property
|
| 153 |
+
def name(self) -> str:
|
| 154 |
+
return self._name
|
| 155 |
+
|
| 156 |
+
@property
|
| 157 |
+
def model(self) -> str:
|
| 158 |
+
return self._model
|
| 159 |
+
|
| 160 |
+
def _resolve_api_key(self) -> str:
|
| 161 |
+
"""Résout la clé API : explicite > env var.
|
| 162 |
+
|
| 163 |
+
Lève ``OCRAdapterError`` si aucune clé n'est disponible.
|
| 164 |
+
"""
|
| 165 |
+
key = self._explicit_api_key or os.environ.get("MISTRAL_API_KEY")
|
| 166 |
+
if not key:
|
| 167 |
+
raise OCRAdapterError(
|
| 168 |
+
f"{self.name} : clé API Mistral manquante. "
|
| 169 |
+
"Définir MISTRAL_API_KEY ou passer api_key= au "
|
| 170 |
+
"constructeur.",
|
| 171 |
+
)
|
| 172 |
+
return key
|
| 173 |
+
|
| 174 |
+
def _encode_image(self, image_path: Path) -> str:
|
| 175 |
+
"""Retourne ``data:<mime>;base64,<...>`` pour l'image."""
|
| 176 |
+
suffix = image_path.suffix.lower()
|
| 177 |
+
media_type = _MEDIA_TYPES.get(suffix, "image/jpeg")
|
| 178 |
+
image_b64 = base64.b64encode(image_path.read_bytes()).decode("ascii")
|
| 179 |
+
return f"data:{media_type};base64,{image_b64}"
|
| 180 |
+
|
| 181 |
+
def execute(
|
| 182 |
+
self,
|
| 183 |
+
inputs: dict[ArtifactType, Artifact],
|
| 184 |
+
params: dict[str, Any],
|
| 185 |
+
context: Any,
|
| 186 |
+
) -> dict[ArtifactType, Artifact]:
|
| 187 |
+
"""Exécute Mistral OCR sur l'image fournie.
|
| 188 |
+
|
| 189 |
+
Route vers l'API appropriée selon ``self.model`` :
|
| 190 |
+
- ``mistral-ocr-*`` → ``/v1/ocr`` via ``urllib`` ;
|
| 191 |
+
- ``pixtral-*`` → API chat/vision via SDK ``mistralai``.
|
| 192 |
+
|
| 193 |
+
Raises
|
| 194 |
+
------
|
| 195 |
+
OCRAdapterError
|
| 196 |
+
Erreur d'input, clé manquante, SDK absent (pour pixtral),
|
| 197 |
+
ou API Mistral en erreur.
|
| 198 |
+
"""
|
| 199 |
+
if ArtifactType.IMAGE not in inputs:
|
| 200 |
+
raise OCRAdapterError(
|
| 201 |
+
f"{self.name} : input IMAGE manquant.",
|
| 202 |
+
)
|
| 203 |
+
image_artifact = inputs[ArtifactType.IMAGE]
|
| 204 |
+
if image_artifact.uri is None:
|
| 205 |
+
raise OCRAdapterError(
|
| 206 |
+
f"{self.name} : artefact image "
|
| 207 |
+
f"{image_artifact.id!r} sans URI.",
|
| 208 |
+
)
|
| 209 |
+
image_path = Path(image_artifact.uri)
|
| 210 |
+
if not image_path.exists():
|
| 211 |
+
raise OCRAdapterError(
|
| 212 |
+
f"{self.name} : image introuvable {image_path!r}.",
|
| 213 |
+
)
|
| 214 |
+
|
| 215 |
+
api_key = self._resolve_api_key()
|
| 216 |
+
image_url = self._encode_image(image_path)
|
| 217 |
+
|
| 218 |
+
if "mistral-ocr" in self._model.lower():
|
| 219 |
+
text = self._call_native_ocr_api(image_url, api_key)
|
| 220 |
+
else:
|
| 221 |
+
text = self._call_chat_vision_api(image_url, api_key)
|
| 222 |
+
|
| 223 |
+
text_path = (
|
| 224 |
+
image_path.parent / f"{image_path.stem}.{self.name}.txt"
|
| 225 |
+
)
|
| 226 |
+
text_path.write_text(text, encoding="utf-8")
|
| 227 |
+
|
| 228 |
+
return {
|
| 229 |
+
ArtifactType.RAW_TEXT: Artifact(
|
| 230 |
+
id=f"{context.document_id}:{self.name}:raw_text",
|
| 231 |
+
document_id=context.document_id,
|
| 232 |
+
type=ArtifactType.RAW_TEXT,
|
| 233 |
+
produced_by_step="ocr",
|
| 234 |
+
uri=str(text_path),
|
| 235 |
+
),
|
| 236 |
+
}
|
| 237 |
+
|
| 238 |
+
# ──────────────────────────────────────────────────────────────
|
| 239 |
+
# API natives
|
| 240 |
+
# ──────────────────────────────────────────────────────────────
|
| 241 |
+
|
| 242 |
+
def _call_native_ocr_api(self, image_url: str, api_key: str) -> str:
|
| 243 |
+
"""Appelle ``POST /v1/ocr`` via urllib et retourne le markdown
|
| 244 |
+
concaténé."""
|
| 245 |
+
payload = json.dumps({
|
| 246 |
+
"model": self._model,
|
| 247 |
+
"document": {"type": "image_url", "image_url": image_url},
|
| 248 |
+
}).encode("utf-8")
|
| 249 |
+
req = urllib.request.Request(
|
| 250 |
+
"https://api.mistral.ai/v1/ocr",
|
| 251 |
+
data=payload,
|
| 252 |
+
headers={
|
| 253 |
+
"Authorization": f"Bearer {api_key}",
|
| 254 |
+
"Content-Type": "application/json",
|
| 255 |
+
},
|
| 256 |
+
method="POST",
|
| 257 |
+
)
|
| 258 |
+
try:
|
| 259 |
+
with urllib.request.urlopen(req, timeout=self._timeout) as resp:
|
| 260 |
+
data = json.loads(resp.read().decode())
|
| 261 |
+
except Exception as exc:
|
| 262 |
+
raise OCRAdapterError(
|
| 263 |
+
f"{self.name} : erreur API Mistral /v1/ocr : "
|
| 264 |
+
f"{type(exc).__name__}: {exc}",
|
| 265 |
+
) from exc
|
| 266 |
+
pages = data.get("pages", [])
|
| 267 |
+
text = "\n\n".join(p.get("markdown", "") for p in pages).strip()
|
| 268 |
+
return text
|
| 269 |
+
|
| 270 |
+
def _call_chat_vision_api(self, image_url: str, api_key: str) -> str:
|
| 271 |
+
"""Appelle l'API chat/vision Mistral via le SDK ``mistralai``."""
|
| 272 |
+
try:
|
| 273 |
+
try:
|
| 274 |
+
from mistralai.client import Mistral
|
| 275 |
+
except ImportError:
|
| 276 |
+
from mistralai import Mistral # type: ignore[no-redef]
|
| 277 |
+
except ImportError as exc:
|
| 278 |
+
raise OCRAdapterError(
|
| 279 |
+
f"{self.name} : SDK 'mistralai' non installé. "
|
| 280 |
+
"Installer avec : pip install mistralai",
|
| 281 |
+
) from exc
|
| 282 |
+
|
| 283 |
+
client = Mistral(api_key=api_key)
|
| 284 |
+
try:
|
| 285 |
+
response = client.chat.complete(
|
| 286 |
+
model=self._model,
|
| 287 |
+
messages=[
|
| 288 |
+
{
|
| 289 |
+
"role": "user",
|
| 290 |
+
"content": [
|
| 291 |
+
{"type": "text", "text": self._prompt},
|
| 292 |
+
{"type": "image_url", "image_url": image_url},
|
| 293 |
+
],
|
| 294 |
+
},
|
| 295 |
+
],
|
| 296 |
+
max_tokens=self._max_tokens,
|
| 297 |
+
)
|
| 298 |
+
except Exception as exc:
|
| 299 |
+
raise OCRAdapterError(
|
| 300 |
+
f"{self.name} : erreur API Mistral chat : "
|
| 301 |
+
f"{type(exc).__name__}: {exc}",
|
| 302 |
+
) from exc
|
| 303 |
+
|
| 304 |
+
try:
|
| 305 |
+
return response.choices[0].message.content or ""
|
| 306 |
+
except (AttributeError, IndexError) as exc:
|
| 307 |
+
raise OCRAdapterError(
|
| 308 |
+
f"{self.name} : réponse Mistral chat malformée : {exc}",
|
| 309 |
+
) from exc
|
| 310 |
+
|
| 311 |
+
|
| 312 |
+
__all__ = ["MistralOCRAdapter"]
|
|
@@ -0,0 +1,390 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Sprint A14-S32 — ``MistralOCRAdapter`` natif au contrat S26."""
|
| 2 |
+
|
| 3 |
+
from __future__ import annotations
|
| 4 |
+
|
| 5 |
+
import sys
|
| 6 |
+
from pathlib import Path
|
| 7 |
+
from unittest.mock import MagicMock, patch
|
| 8 |
+
|
| 9 |
+
import pytest
|
| 10 |
+
|
| 11 |
+
from picarones.adapters.ocr import (
|
| 12 |
+
BaseOCRAdapter,
|
| 13 |
+
MistralOCRAdapter,
|
| 14 |
+
OCRAdapterError,
|
| 15 |
+
)
|
| 16 |
+
from picarones.domain.artifacts import Artifact, ArtifactType
|
| 17 |
+
from picarones.pipeline.types import RunContext
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
def _make_image_artifact(uri: str) -> Artifact:
|
| 21 |
+
return Artifact(
|
| 22 |
+
id="d1:initial:image",
|
| 23 |
+
document_id="d1",
|
| 24 |
+
type=ArtifactType.IMAGE,
|
| 25 |
+
uri=uri,
|
| 26 |
+
)
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
def _make_context() -> RunContext:
|
| 30 |
+
return RunContext(
|
| 31 |
+
document_id="d1",
|
| 32 |
+
code_version="1.0.0",
|
| 33 |
+
pipeline_name="test",
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
def _make_dummy_image(tmp_path: Path) -> Path:
|
| 38 |
+
path = tmp_path / "page.png"
|
| 39 |
+
path.write_bytes(b"\x89PNG\r\n\x1a\nfakeimagebytes")
|
| 40 |
+
return path
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 44 |
+
# Constructeur
|
| 45 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
class TestMistralOCRAdapterConstructor:
|
| 49 |
+
def test_defaults(self) -> None:
|
| 50 |
+
adapter = MistralOCRAdapter()
|
| 51 |
+
assert adapter.name == "mistral_ocr"
|
| 52 |
+
assert adapter.model == "mistral-ocr-latest"
|
| 53 |
+
|
| 54 |
+
def test_custom_name(self) -> None:
|
| 55 |
+
adapter = MistralOCRAdapter(name="my_mistral")
|
| 56 |
+
assert adapter.name == "my_mistral"
|
| 57 |
+
|
| 58 |
+
def test_custom_model(self) -> None:
|
| 59 |
+
adapter = MistralOCRAdapter(model="pixtral-12b-2409")
|
| 60 |
+
assert adapter.model == "pixtral-12b-2409"
|
| 61 |
+
|
| 62 |
+
def test_rejects_empty_name(self) -> None:
|
| 63 |
+
with pytest.raises(OCRAdapterError, match="vide"):
|
| 64 |
+
MistralOCRAdapter(name="")
|
| 65 |
+
|
| 66 |
+
def test_rejects_invalid_chars_in_name(self) -> None:
|
| 67 |
+
with pytest.raises(OCRAdapterError, match="invalide"):
|
| 68 |
+
MistralOCRAdapter(name="bad name")
|
| 69 |
+
|
| 70 |
+
def test_rejects_non_positive_max_tokens(self) -> None:
|
| 71 |
+
with pytest.raises(OCRAdapterError, match="max_tokens"):
|
| 72 |
+
MistralOCRAdapter(max_tokens=0)
|
| 73 |
+
with pytest.raises(OCRAdapterError, match="max_tokens"):
|
| 74 |
+
MistralOCRAdapter(max_tokens=-1)
|
| 75 |
+
|
| 76 |
+
def test_rejects_non_positive_timeout(self) -> None:
|
| 77 |
+
with pytest.raises(OCRAdapterError, match="timeout_seconds"):
|
| 78 |
+
MistralOCRAdapter(timeout_seconds=0)
|
| 79 |
+
with pytest.raises(OCRAdapterError, match="timeout_seconds"):
|
| 80 |
+
MistralOCRAdapter(timeout_seconds=-1.0)
|
| 81 |
+
|
| 82 |
+
|
| 83 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 84 |
+
# Contrat BaseOCRAdapter
|
| 85 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 86 |
+
|
| 87 |
+
|
| 88 |
+
class TestMistralOCRAdapterContract:
|
| 89 |
+
def test_inherits_base_adapter(self) -> None:
|
| 90 |
+
adapter = MistralOCRAdapter()
|
| 91 |
+
assert isinstance(adapter, BaseOCRAdapter)
|
| 92 |
+
|
| 93 |
+
def test_input_types(self) -> None:
|
| 94 |
+
assert MistralOCRAdapter.input_types == frozenset({ArtifactType.IMAGE})
|
| 95 |
+
|
| 96 |
+
def test_output_types(self) -> None:
|
| 97 |
+
assert MistralOCRAdapter.output_types == frozenset({ArtifactType.RAW_TEXT})
|
| 98 |
+
|
| 99 |
+
def test_execution_mode_is_io(self) -> None:
|
| 100 |
+
"""Mistral OCR fait des appels HTTP — IO-bound, ThreadPool."""
|
| 101 |
+
assert MistralOCRAdapter.execution_mode == "io"
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 105 |
+
# API key resolution
|
| 106 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 107 |
+
|
| 108 |
+
|
| 109 |
+
class TestMistralOCRApiKey:
|
| 110 |
+
def test_explicit_key_takes_priority(self) -> None:
|
| 111 |
+
adapter = MistralOCRAdapter(api_key="explicit_key")
|
| 112 |
+
# Mock l'env pour s'assurer qu'on n'utilise pas la valeur env.
|
| 113 |
+
with patch.dict("os.environ", {"MISTRAL_API_KEY": "env_key"}):
|
| 114 |
+
assert adapter._resolve_api_key() == "explicit_key"
|
| 115 |
+
|
| 116 |
+
def test_env_key_used_when_no_explicit(self) -> None:
|
| 117 |
+
adapter = MistralOCRAdapter()
|
| 118 |
+
with patch.dict("os.environ", {"MISTRAL_API_KEY": "env_key"}):
|
| 119 |
+
assert adapter._resolve_api_key() == "env_key"
|
| 120 |
+
|
| 121 |
+
def test_no_key_raises(self) -> None:
|
| 122 |
+
adapter = MistralOCRAdapter()
|
| 123 |
+
# Vide l'env de MISTRAL_API_KEY.
|
| 124 |
+
with patch.dict("os.environ", {}, clear=True):
|
| 125 |
+
with pytest.raises(OCRAdapterError, match="MISTRAL_API_KEY"):
|
| 126 |
+
adapter._resolve_api_key()
|
| 127 |
+
|
| 128 |
+
|
| 129 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 130 |
+
# Encoding
|
| 131 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
class TestMistralOCREncoding:
|
| 135 |
+
def test_png_extension_yields_png_mime(self, tmp_path: Path) -> None:
|
| 136 |
+
adapter = MistralOCRAdapter()
|
| 137 |
+
image_path = _make_dummy_image(tmp_path)
|
| 138 |
+
encoded = adapter._encode_image(image_path)
|
| 139 |
+
assert encoded.startswith("data:image/png;base64,")
|
| 140 |
+
|
| 141 |
+
def test_jpg_extension_yields_jpeg_mime(self, tmp_path: Path) -> None:
|
| 142 |
+
adapter = MistralOCRAdapter()
|
| 143 |
+
path = tmp_path / "img.jpg"
|
| 144 |
+
path.write_bytes(b"jpegbytes")
|
| 145 |
+
encoded = adapter._encode_image(path)
|
| 146 |
+
assert encoded.startswith("data:image/jpeg;base64,")
|
| 147 |
+
|
| 148 |
+
def test_unknown_extension_defaults_to_jpeg(self, tmp_path: Path) -> None:
|
| 149 |
+
adapter = MistralOCRAdapter()
|
| 150 |
+
path = tmp_path / "img.xyz"
|
| 151 |
+
path.write_bytes(b"random")
|
| 152 |
+
encoded = adapter._encode_image(path)
|
| 153 |
+
assert encoded.startswith("data:image/jpeg;base64,")
|
| 154 |
+
|
| 155 |
+
|
| 156 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 157 |
+
# Input validation
|
| 158 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 159 |
+
|
| 160 |
+
|
| 161 |
+
class TestMistralOCRInputValidation:
|
| 162 |
+
def test_missing_image_input_raises(self) -> None:
|
| 163 |
+
adapter = MistralOCRAdapter(api_key="x")
|
| 164 |
+
with pytest.raises(OCRAdapterError, match="IMAGE manquant"):
|
| 165 |
+
adapter.execute(inputs={}, params={}, context=_make_context())
|
| 166 |
+
|
| 167 |
+
def test_image_artifact_without_uri_raises(self) -> None:
|
| 168 |
+
adapter = MistralOCRAdapter(api_key="x")
|
| 169 |
+
artifact = Artifact(
|
| 170 |
+
id="d1:img",
|
| 171 |
+
document_id="d1",
|
| 172 |
+
type=ArtifactType.IMAGE,
|
| 173 |
+
uri=None,
|
| 174 |
+
)
|
| 175 |
+
with pytest.raises(OCRAdapterError, match="sans URI"):
|
| 176 |
+
adapter.execute(
|
| 177 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 178 |
+
params={},
|
| 179 |
+
context=_make_context(),
|
| 180 |
+
)
|
| 181 |
+
|
| 182 |
+
def test_image_path_does_not_exist_raises(self) -> None:
|
| 183 |
+
adapter = MistralOCRAdapter(api_key="x")
|
| 184 |
+
artifact = _make_image_artifact("/nonexistent/img.png")
|
| 185 |
+
with pytest.raises(OCRAdapterError, match="introuvable"):
|
| 186 |
+
adapter.execute(
|
| 187 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 188 |
+
params={},
|
| 189 |
+
context=_make_context(),
|
| 190 |
+
)
|
| 191 |
+
|
| 192 |
+
def test_no_api_key_raises(self, tmp_path: Path) -> None:
|
| 193 |
+
adapter = MistralOCRAdapter() # pas d'api_key explicite
|
| 194 |
+
image_path = _make_dummy_image(tmp_path)
|
| 195 |
+
artifact = _make_image_artifact(str(image_path))
|
| 196 |
+
with patch.dict("os.environ", {}, clear=True):
|
| 197 |
+
with pytest.raises(OCRAdapterError, match="MISTRAL_API_KEY"):
|
| 198 |
+
adapter.execute(
|
| 199 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 200 |
+
params={},
|
| 201 |
+
context=_make_context(),
|
| 202 |
+
)
|
| 203 |
+
|
| 204 |
+
|
| 205 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 206 |
+
# /v1/ocr API (mistral-ocr-* models)
|
| 207 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 208 |
+
|
| 209 |
+
|
| 210 |
+
class TestMistralOCRNativeAPI:
|
| 211 |
+
def _mock_urlopen_ok(self, response_json: dict):
|
| 212 |
+
"""Helper : retourne un context manager qui mock urlopen."""
|
| 213 |
+
mock_resp = MagicMock()
|
| 214 |
+
mock_resp.read.return_value = repr(response_json).encode()
|
| 215 |
+
# On ne peut pas json.dumps un dict avec json.dumps directement
|
| 216 |
+
# à cause du repr ; on encode proprement.
|
| 217 |
+
import json as _json
|
| 218 |
+
mock_resp.read.return_value = _json.dumps(response_json).encode()
|
| 219 |
+
mock_resp.__enter__.return_value = mock_resp
|
| 220 |
+
return patch("urllib.request.urlopen", return_value=mock_resp)
|
| 221 |
+
|
| 222 |
+
def test_native_api_concatenates_pages(self, tmp_path: Path) -> None:
|
| 223 |
+
adapter = MistralOCRAdapter(api_key="x")
|
| 224 |
+
image_path = _make_dummy_image(tmp_path)
|
| 225 |
+
artifact = _make_image_artifact(str(image_path))
|
| 226 |
+
|
| 227 |
+
response_json = {
|
| 228 |
+
"pages": [
|
| 229 |
+
{"markdown": "Page 1 contenu"},
|
| 230 |
+
{"markdown": "Page 2 contenu"},
|
| 231 |
+
],
|
| 232 |
+
}
|
| 233 |
+
|
| 234 |
+
with self._mock_urlopen_ok(response_json):
|
| 235 |
+
result = adapter.execute(
|
| 236 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 237 |
+
params={},
|
| 238 |
+
context=_make_context(),
|
| 239 |
+
)
|
| 240 |
+
out_text = Path(result[ArtifactType.RAW_TEXT].uri).read_text(
|
| 241 |
+
encoding="utf-8",
|
| 242 |
+
)
|
| 243 |
+
assert out_text == "Page 1 contenu\n\nPage 2 contenu"
|
| 244 |
+
|
| 245 |
+
def test_native_api_writes_to_stem_name_pattern(self, tmp_path: Path) -> None:
|
| 246 |
+
adapter = MistralOCRAdapter(api_key="x", name="my_mistral")
|
| 247 |
+
image_path = _make_dummy_image(tmp_path)
|
| 248 |
+
artifact = _make_image_artifact(str(image_path))
|
| 249 |
+
|
| 250 |
+
with self._mock_urlopen_ok({"pages": [{"markdown": "x"}]}):
|
| 251 |
+
result = adapter.execute(
|
| 252 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 253 |
+
params={},
|
| 254 |
+
context=_make_context(),
|
| 255 |
+
)
|
| 256 |
+
out_path = Path(result[ArtifactType.RAW_TEXT].uri)
|
| 257 |
+
assert out_path.name == "page.my_mistral.txt"
|
| 258 |
+
|
| 259 |
+
def test_native_api_raises_on_http_error(self, tmp_path: Path) -> None:
|
| 260 |
+
adapter = MistralOCRAdapter(api_key="x")
|
| 261 |
+
image_path = _make_dummy_image(tmp_path)
|
| 262 |
+
artifact = _make_image_artifact(str(image_path))
|
| 263 |
+
|
| 264 |
+
with patch(
|
| 265 |
+
"urllib.request.urlopen",
|
| 266 |
+
side_effect=ConnectionError("API down"),
|
| 267 |
+
):
|
| 268 |
+
with pytest.raises(OCRAdapterError, match="ConnectionError"):
|
| 269 |
+
adapter.execute(
|
| 270 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 271 |
+
params={},
|
| 272 |
+
context=_make_context(),
|
| 273 |
+
)
|
| 274 |
+
|
| 275 |
+
|
| 276 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 277 |
+
# Vision/chat API (pixtral-* models)
|
| 278 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 279 |
+
|
| 280 |
+
|
| 281 |
+
class TestMistralOCRVisionAPI:
|
| 282 |
+
def test_pixtral_routes_to_vision_api(self, tmp_path: Path) -> None:
|
| 283 |
+
adapter = MistralOCRAdapter(
|
| 284 |
+
api_key="x",
|
| 285 |
+
model="pixtral-12b-2409",
|
| 286 |
+
)
|
| 287 |
+
image_path = _make_dummy_image(tmp_path)
|
| 288 |
+
artifact = _make_image_artifact(str(image_path))
|
| 289 |
+
|
| 290 |
+
# Mock le SDK mistralai.
|
| 291 |
+
mock_message = MagicMock()
|
| 292 |
+
mock_message.content = "Texte transcrit par pixtral."
|
| 293 |
+
mock_choice = MagicMock(message=mock_message)
|
| 294 |
+
mock_response = MagicMock()
|
| 295 |
+
mock_response.choices = [mock_choice]
|
| 296 |
+
|
| 297 |
+
mock_client = MagicMock()
|
| 298 |
+
mock_client.chat.complete.return_value = mock_response
|
| 299 |
+
|
| 300 |
+
fake_module = MagicMock()
|
| 301 |
+
fake_module.Mistral = MagicMock(return_value=mock_client)
|
| 302 |
+
fake_client_module = MagicMock()
|
| 303 |
+
fake_client_module.Mistral = fake_module.Mistral
|
| 304 |
+
|
| 305 |
+
with patch.dict(sys.modules, {
|
| 306 |
+
"mistralai": fake_module,
|
| 307 |
+
"mistralai.client": fake_client_module,
|
| 308 |
+
}):
|
| 309 |
+
result = adapter.execute(
|
| 310 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 311 |
+
params={},
|
| 312 |
+
context=_make_context(),
|
| 313 |
+
)
|
| 314 |
+
|
| 315 |
+
out_text = Path(result[ArtifactType.RAW_TEXT].uri).read_text(
|
| 316 |
+
encoding="utf-8",
|
| 317 |
+
)
|
| 318 |
+
assert out_text == "Texte transcrit par pixtral."
|
| 319 |
+
|
| 320 |
+
def test_pixtral_sdk_missing_raises_clean_error(
|
| 321 |
+
self, tmp_path: Path,
|
| 322 |
+
) -> None:
|
| 323 |
+
adapter = MistralOCRAdapter(api_key="x", model="pixtral-12b")
|
| 324 |
+
image_path = _make_dummy_image(tmp_path)
|
| 325 |
+
artifact = _make_image_artifact(str(image_path))
|
| 326 |
+
|
| 327 |
+
with patch.dict(sys.modules, {
|
| 328 |
+
"mistralai": None,
|
| 329 |
+
"mistralai.client": None,
|
| 330 |
+
}):
|
| 331 |
+
with pytest.raises(OCRAdapterError, match="mistralai"):
|
| 332 |
+
adapter.execute(
|
| 333 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 334 |
+
params={},
|
| 335 |
+
context=_make_context(),
|
| 336 |
+
)
|
| 337 |
+
|
| 338 |
+
def test_pixtral_api_error_wrapped(self, tmp_path: Path) -> None:
|
| 339 |
+
adapter = MistralOCRAdapter(api_key="x", model="pixtral-12b")
|
| 340 |
+
image_path = _make_dummy_image(tmp_path)
|
| 341 |
+
artifact = _make_image_artifact(str(image_path))
|
| 342 |
+
|
| 343 |
+
mock_client = MagicMock()
|
| 344 |
+
mock_client.chat.complete.side_effect = RuntimeError("API error")
|
| 345 |
+
|
| 346 |
+
fake_module = MagicMock()
|
| 347 |
+
fake_module.Mistral = MagicMock(return_value=mock_client)
|
| 348 |
+
fake_client_module = MagicMock()
|
| 349 |
+
fake_client_module.Mistral = fake_module.Mistral
|
| 350 |
+
|
| 351 |
+
with patch.dict(sys.modules, {
|
| 352 |
+
"mistralai": fake_module,
|
| 353 |
+
"mistralai.client": fake_client_module,
|
| 354 |
+
}):
|
| 355 |
+
with pytest.raises(OCRAdapterError, match="RuntimeError.*API error"):
|
| 356 |
+
adapter.execute(
|
| 357 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 358 |
+
params={},
|
| 359 |
+
context=_make_context(),
|
| 360 |
+
)
|
| 361 |
+
|
| 362 |
+
|
| 363 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 364 |
+
# Artifact ID
|
| 365 |
+
# ──────────────────────────────────────────────────────────────────────
|
| 366 |
+
|
| 367 |
+
|
| 368 |
+
class TestMistralOCRArtifactID:
|
| 369 |
+
def test_artifact_id_uses_adapter_name(self, tmp_path: Path) -> None:
|
| 370 |
+
adapter = MistralOCRAdapter(api_key="x", name="custom")
|
| 371 |
+
image_path = _make_dummy_image(tmp_path)
|
| 372 |
+
artifact = _make_image_artifact(str(image_path))
|
| 373 |
+
|
| 374 |
+
mock_resp = MagicMock()
|
| 375 |
+
import json as _json
|
| 376 |
+
mock_resp.read.return_value = _json.dumps(
|
| 377 |
+
{"pages": [{"markdown": "x"}]},
|
| 378 |
+
).encode()
|
| 379 |
+
mock_resp.__enter__.return_value = mock_resp
|
| 380 |
+
|
| 381 |
+
with patch("urllib.request.urlopen", return_value=mock_resp):
|
| 382 |
+
result = adapter.execute(
|
| 383 |
+
inputs={ArtifactType.IMAGE: artifact},
|
| 384 |
+
params={},
|
| 385 |
+
context=_make_context(),
|
| 386 |
+
)
|
| 387 |
+
produced = result[ArtifactType.RAW_TEXT]
|
| 388 |
+
assert produced.id == "d1:custom:raw_text"
|
| 389 |
+
assert produced.document_id == "d1"
|
| 390 |
+
assert produced.produced_by_step == "ocr"
|