Removing DocumentLanguage, as both System Context and User Context are to be defined on DocumentVersion level.

Finetuning of embedding workers.
This commit is contained in:
Josako
2024-06-06 15:26:49 +02:00
parent 1a25313673
commit 27b6de8734
21 changed files with 301 additions and 295 deletions

View File

@@ -1,24 +0,0 @@
{% extends 'base.html' %}
{% from 'macros.html' import render_selectable_table, render_pagination %}
{% block title %}Document Languages{% endblock %}
{% block content_title %}Document Languages{% endblock %}
{% block content_description %}View Languages for {{ document }}{% endblock %}
{% block content_class %}<div class="col-xl-12 col-lg-5 col-md-7 mx-auto">{% endblock %}
{% block content %}
<div class="container">
<form method="POST" action="{{ url_for('document_bp.handle_document_language_selection') }}">
{{ render_selectable_table(headers=["Document Language ID", "Language", "User Context", "System Context"], rows=rows, selectable=True, id="documentsTable") }}
<div class="form-group mt-3">
<button type="submit" name="action" value="edit_document_language" class="btn btn-primary">Edit Document Language</button>
<button type="submit" name="action" value="document_versions" class="btn btn-secondary">Show Document Versions</button>
</div>
</form>
</div>
{% endblock %}
{% block content_footer %}
{{ render_pagination(pagination, 'document_bp.documents') }}
{% endblock %}

View File

@@ -12,7 +12,8 @@
<form method="POST" action="{{ url_for('document_bp.handle_document_version_selection') }}">
{{ render_selectable_table(headers=["Document Version ID", "URL", "File Location", "File Name", "File Type", "Processing", "Processing Start", "Proceeing Finish"], rows=rows, selectable=True, id="versionsTable") }}
<div class="form-group mt-3">
<button type="submit" name="action" value="process_document_version" class="btn btn-primary">Process Document Version</button>
<button type="submit" name="action" value="edit_document_version" class="btn btn-primary">Edit Document Version</button>
<button type="submit" name="action" value="process_document_version" class="btn btn-secondary">Process Document Version</button>
</div>
</form>
</div>

View File

@@ -13,7 +13,7 @@
{{ render_selectable_table(headers=["Document ID", "Name", "Valid From", "Valid To"], rows=rows, selectable=True, id="documentsTable") }}
<div class="form-group mt-3">
<button type="submit" name="action" value="edit_document" class="btn btn-primary">Edit Document</button>
<button type="submit" name="action" value="document_languages" class="btn btn-secondary">Show Document Languages</button>
<button type="submit" name="action" value="document_versions" class="btn btn-secondary">Show Document Versions</button>
</div>
</form>
</div>

View File

@@ -72,6 +72,7 @@
{'name': 'Tenant List', 'url': '/user/select_tenant', 'roles': ['Super User']},
{'name': 'Tenant Registration', 'url': '/user/tenant', 'roles': ['Super User']},
{'name': 'Tenant Overview', 'url': '/user/tenant_overview', 'roles': ['Super User', 'Tenant Admin']},
{'name': 'Edit Tenant', 'url': '/user/tenant/' ~ session['tenant'].get('id'), 'roles': ['Super User', 'Tenant Admin']},
{'name': 'Tenant Domains', 'url': '/user/view_tenant_domains', 'roles': ['Super User', 'Tenant Admin']},
{'name': 'Tenant Domain Registration', 'url': '/user/tenant_domain', 'roles': ['Super User', 'Tenant Admin']},
{'name': 'User List', 'url': '/user/view_users', 'roles': ['Super User', 'Tenant Admin']},

View File

@@ -40,6 +40,11 @@
Embedding Search
</a>
</li>
<li class="nav-item">
<a class="nav-link mb-0 px-0 py-1" data-toggle="tab" href="#tuning-tab" role="tab" aria-controls="html-chunking" aria-selected="false">
Tuning
</a>
</li>
</ul>
</div>
<div class="tab-content tab-space">
@@ -73,6 +78,13 @@
{{ render_included_field(field, disabled_fields=es_fields, include_fields=es_fields) }}
{% endfor %}
</div>
<!-- Tuning Settings Tab -->
<div class="tab-pane fade" id="tuning-tab" role="tabpanel">
{% set tuning_fields = ['embed_tuning', 'rag_tuning', ] %}
{% for field in form %}
{{ render_included_field(field, disabled_fields=tuning_fields, include_fields=tuning_fields) }}
{% endfor %}
</div>
</div>
</div>
</div>

View File

@@ -47,7 +47,7 @@ class EditDocumentForm(FlaskForm):
submit = SubmitField('Submit')
class EditDocumentLanguageForm(FlaskForm):
class EditDocumentVersionForm(FlaskForm):
language = StringField('Language')
user_context = TextAreaField('User Context', validators=[Optional()])
system_context = TextAreaField('System Context', validators=[Optional()])

View File

@@ -13,9 +13,9 @@ from requests.exceptions import SSLError
from urllib.parse import urlparse
import io
from common.models.document import Document, DocumentLanguage, DocumentVersion
from common.models.document import Document, DocumentVersion
from common.extensions import db
from .document_forms import AddDocumentForm, AddURLForm, EditDocumentForm, EditDocumentLanguageForm
from .document_forms import AddDocumentForm, AddURLForm, EditDocumentForm, EditDocumentVersionForm
from common.utils.middleware import mw_before_request
from common.utils.celery_utils import current_celery
from common.utils.nginx_utils import prefixed_url_for
@@ -59,7 +59,7 @@ def add_document():
filename = secure_filename(file.filename)
extension = filename.rsplit('.', 1)[1].lower()
new_doc, new_doc_lang, new_doc_vers = create_document_stack(form, file, filename, extension)
new_doc, new_doc_vers = create_document_stack(form, file, filename, extension)
task = current_celery.send_task('create_embeddings', queue='embeddings', args=[
session['tenant']['id'],
@@ -100,7 +100,7 @@ def add_url():
filename += '.html'
extension = 'html'
new_doc, new_doc_lang, new_doc_vers = create_document_stack(form, file, filename, extension)
new_doc, new_doc_vers = create_document_stack(form, file, filename, extension)
task = current_celery.send_task('create_embeddings', queue='embeddings', args=[
session['tenant']['id'],
@@ -146,8 +146,8 @@ def handle_document_selection():
match action:
case 'edit_document':
return redirect(prefixed_url_for('document_bp.edit_document', document_id=doc_id))
case 'document_languages':
return redirect(prefixed_url_for('document_bp.document_languages', document_id=doc_id))
case 'document_versions':
return redirect(prefixed_url_for('document_bp.document_versions', document_id=doc_id))
# Add more conditions for other actions
return redirect(prefixed_url_for('document_bp.documents'))
@@ -180,81 +180,46 @@ def edit_document(document_id):
return render_template('document/edit_document.html', form=form, document_id=document_id)
@document_bp.route('/document_languages/<int:document_id>', methods=['GET', 'POST'])
@document_bp.route('/edit_document_version/<int:document_version_id>', methods=['GET', 'POST'])
@roles_accepted('Super User', 'Tenant Admin')
def document_languages(document_id):
doc = Document.query.get_or_404(document_id)
doc_desc = f'Document {doc.id}: {doc.name}'
page = request.args.get('page', 1, type=int)
per_page = request.args.get('per_page', 10, type=int)
query = DocumentLanguage.query.filter_by(document_id=document_id).order_by(DocumentLanguage.language)
pagination = query.paginate(page=page, per_page=per_page, error_out=False)
doc_langs = pagination.items
rows = prepare_table_for_macro(doc_langs, [('id', ''), ('language', ''), ('user_context', ''),
('system_context', '')])
return render_template('document/document_languages.html', rows=rows, pagination=pagination, document=doc_desc)
@document_bp.route('/handle_document_language_selection', methods=['POST'])
@roles_accepted('Super User', 'Tenant Admin')
def handle_document_language_selection():
document_language_identification = request.form['selected_row']
doc_lang_id = ast.literal_eval(document_language_identification).get('value')
action = request.form['action']
match action:
case 'edit_document_language':
return redirect(prefixed_url_for('document_bp.edit_document_language', document_language_id=doc_lang_id))
case 'document_versions':
return redirect(prefixed_url_for('document_bp.document_versions', document_language_id=doc_lang_id))
# Add more conditions for other actions
return redirect(prefixed_url_for('document_bp.document_languages'))
@document_bp.route('/edit_document_language/<int:document_language_id>', methods=['GET', 'POST'])
@roles_accepted('Super User', 'Tenant Admin')
def edit_document_language(document_language_id):
doc_lang = DocumentLanguage.query.get_or_404(document_language_id)
form = EditDocumentLanguageForm(obj=doc_lang)
def edit_document_version(document_version_id):
doc_vers = DocumentVersion.query.get_or_404(document_version_id)
form = EditDocumentVersionForm(obj=doc_vers)
if form.validate_on_submit():
doc_lang.user_context = form.user_context.data
doc_vers.user_context = form.user_context.data
update_logging_information(doc_lang, dt.now(tz.utc))
update_logging_information(doc_vers, dt.now(tz.utc))
try:
db.session.add(doc_lang)
db.session.add(doc_vers)
db.session.commit()
flash(f'Document Language {doc_lang.id} updated successfully', 'success')
flash(f'Document Version {doc_vers.id} updated successfully', 'success')
except SQLAlchemyError as e:
db.session.rollback()
flash(f'Error updating document language: {e}', 'danger')
current_app.logger.error(f'Error updating document language {doc_lang.id} '
flash(f'Error updating document version: {e}', 'danger')
current_app.logger.error(f'Error updating document version {doc_vers.id} '
f'for tenant {session['tenant']['id']}: {e}')
else:
form_validation_failed(request, form)
return render_template('document/edit_document_language.html', form=form, document_langauge_id=document_language_id,
doc_details=f'Document {doc_lang.document.name}')
return render_template('document/edit_document_version.html', form=form, document_version_id=document_version_id,
doc_details=f'Document {doc_vers.document.name}')
@document_bp.route('/document_versions/<int:document_language_id>', methods=['GET', 'POST'])
@document_bp.route('/document_versions/<int:document_id>', methods=['GET', 'POST'])
@roles_accepted('Super User', 'Tenant Admin')
def document_versions(document_language_id):
def document_versions(document_id):
flash(f'Processing documents is a long running process. Please be careful retriggering processing!', 'danger')
doc_lang = DocumentLanguage.query.get_or_404(document_language_id)
doc_desc = f'Document {doc_lang.document.name}, Language {doc_lang.language}'
doc_vers = DocumentVersion.query.get_or_404(document_id)
doc_desc = f'Document {doc_vers.document.name}, Language {doc_vers.language}'
page = request.args.get('page', 1, type=int)
per_page = request.args.get('per_page', 10, type=int)
query = DocumentVersion.query.filter_by(doc_lang_id=document_language_id).order_by(desc(DocumentVersion.id))
query = (DocumentVersion.query.filter_by(doc_id=document_id)
.order_by(DocumentVersion.language)
.order_by(desc(DocumentVersion.id)))
pagination = query.paginate(page=page, per_page=per_page, error_out=False)
doc_langs = pagination.items
@@ -276,6 +241,8 @@ def handle_document_version_selection():
action = request.form['action']
match action:
case 'edit_document_version':
return redirect(prefixed_url_for('document_bp.edit_document_version', document_version_id=doc_vers_id))
case 'process_document_version':
process_version(doc_vers_id)
# Add more conditions for other actions
@@ -315,17 +282,11 @@ def create_document_stack(form, file, filename, extension):
# Create the Document
new_doc = create_document(form, filename)
# Create the DocumentLanguage
new_doc_lang = create_language_for_document(new_doc, form.language.data, form.user_context.data)
# Create the DocumentVersion
new_doc_vers = DocumentVersion()
new_doc_vers.document_language = new_doc_lang
set_logging_information(new_doc_vers, dt.now(tz.utc))
new_doc_vers = create_version_for_document(new_doc, form.language.data, form.user_context.data)
try:
db.session.add(new_doc)
db.session.add(new_doc_lang)
db.session.add(new_doc_vers)
db.session.commit()
except SQLAlchemyError as e:
@@ -338,30 +299,12 @@ def create_document_stack(form, file, filename, extension):
current_app.logger.error('Unknown error')
raise
try:
new_doc_lang = db.session.merge(new_doc_lang)
new_doc_vers = db.session.merge(new_doc_vers)
new_doc_lang.latest_version_id = new_doc_vers.id
db.session.commit()
except SQLAlchemyError as e:
current_app.logger.error(f'Error adding document for tenant {session["tenant"]["id"]}: {e}')
flash('Error adding document.', 'error')
db.session.rollback()
error = e.args
raise
except Exception as e:
current_app.logger.error(f'Error adding document for tenant {session["tenant"]["id"]}: {e}')
flash('Error adding document.', 'error')
db.session.rollback()
error = e.args
raise
current_app.logger.info(f'Document added successfully for tenant {session["tenant"]["id"]}, '
f'Document Version {new_doc.id}')
upload_file_for_version(new_doc_vers, file, extension)
return new_doc, new_doc_lang, new_doc_vers
return new_doc, new_doc_vers
def log_session_state(session, msg=""):
@@ -386,21 +329,21 @@ def create_document(form, filename):
return new_doc
def create_language_for_document(document, language, user_context):
new_doc_lang = DocumentLanguage()
def create_version_for_document(document, language, user_context):
new_doc_vers = DocumentVersion()
if language == '':
new_doc_lang.language = session['default_language']
new_doc_vers.language = session['default_language']
else:
new_doc_lang.language = language
new_doc_vers.language = language
if user_context != '':
new_doc_lang.user_context = user_context
new_doc_vers.user_context = user_context
new_doc_lang.document = document
new_doc_vers.document = document
set_logging_information(new_doc_lang, dt.now(tz.utc))
set_logging_information(new_doc_vers, dt.now(tz.utc))
return new_doc_lang
return new_doc_vers
def upload_file_for_version(doc_vers, file, extension):

View File

@@ -33,6 +33,9 @@ class TenantForm(FlaskForm):
es_similarity_threshold = FloatField('Similarity Threshold for Searching Embeddings (0.5)',
default=0.5,
validators=[NumberRange(min=0, max=1)])
# Tuning variables
embed_tuning = BooleanField('Enable Embedding Tuning', default=False)
rag_tuning = BooleanField('Enable RAG Tuning', default=False)
submit = SubmitField('Submit')

View File

@@ -53,7 +53,9 @@ def tenant():
llm_model=form.llm_model.data,
license_start_date=form.license_start_date.data,
license_end_date=form.license_end_date.data,
allowed_monthly_interactions=form.allowed_monthly_interactions.data)
allowed_monthly_interactions=form.allowed_monthly_interactions.data,
embed_tuning=form.embed_tuning.data,
rag_tuning=form.rag_tuning.data)
# Handle Embedding Variables
new_tenant.html_tags = form.html_tags.data.split(',') if form.html_tags.data else []