Release notes¶
Release 0.4.0 (2026-07-04)¶
LLM updates¶
Cleanup of petro-lingo on latest LLM behaviors for OpenAI API
Migrate from Chat API to the newer Response API
Select the models to expose in petro-lingo for 4.8
Documentation¶
Documentation for translation
Explain confidence mechanism better
Update docs and ZDR models available
Bugs and maintenance¶
Drop the max_input_tokens mechanisms
Enforce confidence unknown/low based on the result
Remove full stop from tooltips
Clean GeoApps-Error for concatenate groups if not the same association
Update petro_lingo so object are being pushed to outgroup
Update to Python 3.12
Relock on pandas ^2.2.1, remove support for python 3.14
Json bug encountered when running petro-lingo relogger
Have consistency in output labels with name, vs data name vs data group name
Gemini crashes on structured output
Use uison directly instead of re-creating dict for tests
Release 0.3.0 (2026-01-12)¶
Widget order and no parallel by default for custom models
Multi-classification for petro-lingo
Add quantitative information to the prompt
Make the relogger available with other models
Change value map names for confidence to high, medium, low
Remove world limitation on LLMs requests
Allow several categories from the same class coexisting in relogger
Change guardrails to structured output
Remove the auto-name mechanism by default “output”
LLM summary tells about quantities while it has no quantitative information as input
Presentation about relogger on real data
Add Author(s) to the case study
translation with LLM
Add a workflow image and a result pie chart image to the Butler presentation
Improve confidence index Chain of Thoughts + form
Clean up the list of offered GPT models - keep the beat/latest only
Adapt petro lingo to Open Router
Improve confidence index Chain of Thoughts + form
With built-in OpenRouter provider, expose a larger subset of models to choose from
Remove adding metadata to target object
Reduce number of confidence bins to low | high
Add translate docs in 4.7
Release 0.2.0 (2025-06-20)¶
Add a function to merge different columns into one text columns
Add a context windows allowing user to add text/select a context file to add to the prompt
Remove duplicates data before sending them to openai to avoid burning tokens
Investigate why context doesn’t seem to influence the answer
LLM cannot deal with spatiality
Structured relogger ready
Create a quantitative test for petro-lingo relogger
Store target score as constant in test file
Test version consistency
Parallel request for petro-lingo
Change ui.jsons according to feedback
Build conda package faster with rattler-build
Petro-lingo (develop) tests fail on Windows
have the version read dynamically in the doc conf.py
Update context windows with new widget
Petro-lingo no key error is not obvious:
Complete package readme and description
Catch no geological before running on all the data
Clean warning from pydantic
Merge application for ask and summarize
Petro-lingo composite bug when one of the association is selected
Improve summary/ask UI layout
Passing empty text raise an error not obvious
Accept coma only as separators
Work the prompt engineering to avoid vague answers
Allow model with Open AI compatibility API
Relock on latest git dev revisions
Add an option to choose the output name
Bring back adjusted ui.json from Analyst
Expose newer OpenAI models for petro-lingo
Composite log separator places header instead
Add a guardrail against quantitative questions + various adjustment
Update petro-lingo doc according to new usage
Backport GA 4.6 - Make the relogger available with other models
Updated env for Analyst 4.6
Update ui.json file as per latest adjustment in Analyst
Release 0.1.0 (2024-06-17)¶
(First release)
New features¶
Bump actions/checkout from 3 to 4
Re-align dependencies to distribute under pro-geology
Improve documentation of petro-lingo
Remove dependency on transformers
Prepare distribution
Precommit config and links to doc, homepage
Fix mypy errors
Merge Release/0.1.0 to main