Third-Party Notices

Used for scratch pretraining and later replay in the current Adriaticum model lineage. The license covers the released dataset or database; underlying web content may remain subject to rights held by its original publishers. Use is also subject to the Common Crawl Terms of Use.

The exact Hugging Face dataset snapshot or revision used at acquisition was not retained in Mijatovic lineage records.

Used for scratch pretraining and later replay in the current Adriaticum model lineage. The license covers the released dataset or database; underlying web content may remain subject to rights held by its original publishers. Use is also subject to the Common Crawl Terms of Use.

The exact Hugging Face dataset snapshot or revision used at acquisition was not retained in Mijatovic lineage records.

Used for scratch pretraining and later replay in the current Adriaticum model lineage. The current official repository record declares CC0 1.0. Because the corpora derive from web crawls, rights in underlying source material may remain with original rightsholders.

Historical acquisition-specific terms were not preserved independently of the current CLARIN.SI repository record.

Used for language adaptation in the current Adriaticum model lineage. Mijatovic selected, normalized, tokenized, and packed portions of the parliamentary corpora for training.

English–Croatian pairs from DGT-TM release 2021, covering the 2020 data update, were selected, normalized, and converted for supervised training in the current Adriaticum model lineage. The European Commission retains ownership of the data.

The database and accompanying software are made available without guarantee. The European Commission is not responsible for loss, injury, or damage arising from reuse; Mijatovic is responsible for its reuse and model outputs.

Selected English conversations were used for supervised training in the current Adriaticum model lineage after language and quality filtering, branch selection, normalization, length filtering, deduplication, and train-validation splitting.

A filtered English subset was used for supervised training in the current Adriaticum model lineage after response selection, quality filtering, normalization, deduplication, and train-validation splitting.

The Serbian-language subset was used for supervised training in the current Adriaticum model lineage after language selection, normalization, validation, deduplication, and train-validation splitting.

A subset of the first-party Identity & Behavior v1 examples inherited by the current Adriaticum model was generated with the DeepSeek Open Platform API in August 2026. Mijatovic subsequently reviewed, filtered, normalized, deduplicated, and validated the material.

This material was AI-generated before Mijatovic processing. Acquisition-time archival copies of the applicable terms were not retained. Current official terms are linked below. DeepSeek does not endorse Mijatovic or Adriaticum.

Used for translation evaluation.

Evaluation only — not used to train the current Adriaticum model. The exact acquired revision and dataset license detail were not preserved in Mijatovic lineage records.

Used as a translation baseline.

Evaluation only — not used to train the current Adriaticum model. The exact acquired model revision and model-artifact license detail were not preserved in Mijatovic lineage records; an associated code repository reports an MIT license.

Selected models were used as comparative baselines.

Evaluation only — not used to train the current Adriaticum model. Exact model identities, revisions, and per-model license detail were not preserved in Mijatovic lineage records.

Used for academic benchmark evaluation.

Evaluation only — not used to train the current Adriaticum model. The exact acquired dataset revision and dataset license detail were not preserved in Mijatovic lineage records.

A subset was consumed by a trained experimental branch.

Used only in an experimental model branch and not present in the current production model lineage. The exact acquired revision was not retained in Mijatovic records.

A subset was consumed by a trained experimental branch.

Used only in an experimental model branch and not present in the current production model lineage. The exact acquired revision was not retained in Mijatovic records.

A subset was consumed by a trained experimental branch.

Used only in an experimental model branch and not present in the current production model lineage. The exact acquired revision was not retained in Mijatovic records.

Interface typeface used by both web applications.

Interface typeface used by the Mijatovic website.

Used to identify a GitHub destination. GitHub and the GitHub logo are trademarks of GitHub, Inc. No affiliation or endorsement is implied. Local variants should follow GitHub trademark and logo guidance where applicable.

Used to identify Mijatovic’s Hugging Face destination. Hugging Face and its logo are trademarks of Hugging Face, Inc. No affiliation or endorsement is implied. Local variants should follow Hugging Face brand guidance where applicable.

Used to identify Mijatovic’s X destination. X and the X logo are trademarks of X Corp. No affiliation or endorsement is implied. Local variants should follow X brand toolkit guidance where applicable.

Cloudflare provides internet-facing edge, security, DNS, and email infrastructure.

Cloudflare Workers are active production infrastructure for Mijatovic web and product surfaces. Cloudflare Worker and container infrastructure is also used for the production server-side Search path.

Supabase provides managed authentication and application data services used in production Mijatovic surfaces.

RunPod Serverless provides production inference hosting for the Adriaticum model. Inference requests are initiated only from Mijatovic server boundaries.