July 2024

2024.7.7

Fix: fix handling of none-like values in Navigator-Fine-Tuning.
Fix: Improve handling of not-nullable zero-values (empty strings, 0-integers) in workflows.
Fix: fix global locales in Tv2 configs.
Fix: prevent recursion error in TabularDP.
Task: Add membership inference attack score to reports.
Task: Grant project access to domain owners for workflows and connections.

Fix: Fix agent resolution of models image.
Fix: Add transform (v1) and classify to new models image.
Task: Improve handling of column types in MySQL and BigQuery connectors.
Task: Set home to /run directory when running hybrid not as root.
Task: Update GPT-x DP fine tuning to use Poisson Sampler.
Task: Add text entity report to Tv2.
Task: Support resolving data from multiple sources in workflow actions.
Task: Support filtering by model_id and/or model_type on /v1/inference/models endpoint.
Task: Add quasi_identifier_count privacy_metrics to synthetic model configs.
Task: Add inference attack score to synthetics reports.

Fix: Properly set default globals and classify configuration values in Tv2.
Fix: Resolve permissions issues running hybrid GPT-x and Tv2 jobs as non-root.
Fix: Validation of BigQuery connector with unspecified dataset.
Fix: Properly recognize valid MSSQL identity column types.
Fix: Fix model-model image resolution for tagged images.
Task: Improve searching of projects via new owned_by query parameter added to /projects endpoint.
Task: Improve error messaging related to JSONL errors in Navigator.

Fix: Hybrid consolidated Model container did not have proper CUDA paths set, causing GPT-x jobs to fail.
Task: Improvements to Tv2 logging. When Tv2 is processing long NER text blobs, ensure that progress is being reported on regular intervals.
Fix: Navigator inference requests would sometimes fail when using the Google Gemini Pro model.
Task: Text SQS update to semantic similarity score. This update is to ensure that the score is penalized for increasing number of synthetic records that are not semantically similar to any training records.
Fix: Properly configure java.nio for Databricks connector.

Task: GPT-x now uses flash attention to speed up training and inference.
Fix: Invalid model configs would sometimes result in 500 errors.
Task: Update to NavFT to ensure rope_scaling_factor is consistent between model training and inference.
Task: Improvements to Navigator prompt templating.
Task: Increase NavFT rope_scaling_factor upper bound to 6 from 2.
Task: Hybrid deployments default to use new consolidated Model and Workflow docker images.

Task: Improvements to Workflow error logging.
Fix: Workflows using Azure Blob Storage would sometimes commit 0 byte blocks causing write failures.

Last updated 1 month ago

Was this helpful?