studio/docs
Archived
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

⚡️(AI) improve formating of ai translation

Browse Source 
The ai translation were quite lossy about formatting.
Colors, background, breaklines, table sizes were
lost in the translation.
We improve the AI translation request to keep
the formatting as close as possible by using
html instead of markdown.
This commit is contained in:
Anthony LC
2025-03-12 10:14:47 +01:00
committed by Anthony LC
parent 9176328200
commit cbf9091d1c
6 changed files with 99 additions and 145 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
src/backend/core/tests/documents/test_api_documents_ai_transform.py
View File

@@ -71,9 +71,8 @@ def test_api_documents_ai_transform_anonymous_success(mock_create):
"""
document = factories.DocumentFactory(link_reach="public", link_role="editor")
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
@@ -83,17 +82,15 @@ def test_api_documents_ai_transform_anonymous_success(mock_create):
assert response.json() == {"answer": "Salut"}
mock_create.assert_called_once_with(
model="llama",
response_format={"type": "json_object"},
messages=[
{
"role": "system",
"content": (
"Summarize the markdown text, preserving language and markdown formatting. "
'Return JSON: {"answer": "your markdown summary"}. Do not provide any other '
"information."
"Do not provide any other information. Preserve the language."
),
},
{"role": "user", "content": '{"markdown_input": "Hello"}'},
{"role": "user", "content": "Hello"},
],
)
@@ -170,9 +167,8 @@ def test_api_documents_ai_transform_authenticated_success(mock_create, reach, ro
document = factories.DocumentFactory(link_reach=reach, link_role=role)
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
@@ -182,16 +178,15 @@ def test_api_documents_ai_transform_authenticated_success(mock_create, reach, ro
assert response.json() == {"answer": "Salut"}
mock_create.assert_called_once_with(
model="llama",
response_format={"type": "json_object"},
messages=[
{
"role": "system",
"content": (
'Answer the prompt in markdown format. Return JSON: {"answer": '
'"Your markdown answer"}. Do not provide any other information.'
"Answer the prompt in markdown format. Preserve the language and markdown "
"formatting. Do not provide any other information. Preserve the language."
),
},
{"role": "user", "content": '{"markdown_input": "Hello"}'},
{"role": "user", "content": "Hello"},
],
)
@@ -246,9 +241,8 @@ def test_api_documents_ai_transform_success(mock_create, via, role, mock_user_te
document=document, team="lasuite", role=role
)
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
@@ -258,16 +252,15 @@ def test_api_documents_ai_transform_success(mock_create, via, role, mock_user_te
assert response.json() == {"answer": "Salut"}
mock_create.assert_called_once_with(
model="llama",
response_format={"type": "json_object"},
messages=[
{
"role": "system",
"content": (
'Answer the prompt in markdown format. Return JSON: {"answer": '
'"Your markdown answer"}. Do not provide any other information.'
"Answer the prompt in markdown format. Preserve the language and markdown "
"formatting. Do not provide any other information. Preserve the language."
),
},
{"role": "user", "content": '{"markdown_input": "Hello"}'},
{"role": "user", "content": "Hello"},
],
)
@@ -315,9 +308,8 @@ def test_api_documents_ai_transform_throttling_document(mock_create):
client = APIClient()
document = factories.DocumentFactory(link_reach="public", link_role="editor")
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-transform/"
@@ -350,9 +342,8 @@ def test_api_documents_ai_transform_throttling_user(mock_create):
client = APIClient()
client.force_login(user)
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
for _ in range(3):

45
src/backend/core/tests/documents/test_api_documents_ai_translate.py
View File

@@ -91,29 +91,28 @@ def test_api_documents_ai_translate_anonymous_success(mock_create):
"""
document = factories.DocumentFactory(link_reach="public", link_role="editor")
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Ola"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
response = APIClient().post(url, {"text": "Hello", "language": "es"})
assert response.status_code == 200
assert response.json() == {"answer": "Salut"}
assert response.json() == {"answer": "Ola"}
mock_create.assert_called_once_with(
model="llama",
response_format={"type": "json_object"},
messages=[
{
"role": "system",
"content": (
"Translate the markdown text to Spanish, preserving markdown formatting. "
'Return JSON: {"answer": "your translated markdown text in Spanish"}. '
"Keep the same html stucture and formatting. "
"Translate the content in the html to the specified language Spanish. "
"Check the translation for accuracy and make any necessary corrections. "
"Do not provide any other information."
),
},
{"role": "user", "content": '{"markdown_input": "Hello"}'},
{"role": "user", "content": "Hello"},
],
)
@@ -190,9 +189,8 @@ def test_api_documents_ai_translate_authenticated_success(mock_create, reach, ro
document = factories.DocumentFactory(link_reach=reach, link_role=role)
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
@@ -202,18 +200,18 @@ def test_api_documents_ai_translate_authenticated_success(mock_create, reach, ro
assert response.json() == {"answer": "Salut"}
mock_create.assert_called_once_with(
model="llama",
response_format={"type": "json_object"},
messages=[
{
"role": "system",
"content": (
"Translate the markdown text to Colombian Spanish, "
"preserving markdown formatting. Return JSON: "
'{"answer": "your translated markdown text in Colombian Spanish"}. '
"Keep the same html stucture and formatting. "
"Translate the content in the html to the "
"specified language Colombian Spanish. "
"Check the translation for accuracy and make any necessary corrections. "
"Do not provide any other information."
),
},
{"role": "user", "content": '{"markdown_input": "Hello"}'},
{"role": "user", "content": "Hello"},
],
)
@@ -268,9 +266,8 @@ def test_api_documents_ai_translate_success(mock_create, via, role, mock_user_te
document=document, team="lasuite", role=role
)
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
@@ -280,18 +277,18 @@ def test_api_documents_ai_translate_success(mock_create, via, role, mock_user_te
assert response.json() == {"answer": "Salut"}
mock_create.assert_called_once_with(
model="llama",
response_format={"type": "json_object"},
messages=[
{
"role": "system",
"content": (
"Translate the markdown text to Colombian Spanish, "
"preserving markdown formatting. Return JSON: "
'{"answer": "your translated markdown text in Colombian Spanish"}. '
"Keep the same html stucture and formatting. "
"Translate the content in the html to the "
"specified language Colombian Spanish. "
"Check the translation for accuracy and make any necessary corrections. "
"Do not provide any other information."
),
},
{"role": "user", "content": '{"markdown_input": "Hello"}'},
{"role": "user", "content": "Hello"},
],
)
@@ -339,9 +336,8 @@ def test_api_documents_ai_translate_throttling_document(mock_create):
client = APIClient()
document = factories.DocumentFactory(link_reach="public", link_role="editor")
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
url = f"/api/v1.0/documents/{document.id!s}/ai-translate/"
@@ -374,9 +370,8 @@ def test_api_documents_ai_translate_throttling_user(mock_create):
client = APIClient()
client.force_login(user)
answer = '{"answer": "Salut"}'
mock_create.return_value = MagicMock(
choices=[MagicMock(message=MagicMock(content=answer))]
choices=[MagicMock(message=MagicMock(content="Salut"))]
)
for _ in range(3):