LLaDA2.2-mini BG + TG statistics

Tokenizer-matched sparse counts for Ruurd/llada2.2-mini. Model/tokenizer: inclusionAI/LLaDA2.2-mini, revision e4f77d25ab5ca86f2223c6cdc6a72b887e20c8a5. Contains no model weights and no raw training documents.

875,507,295 tokens: all 980,000 original response-training documents plus 800M exact tokenizer-specific tokens (720M FineWeb-Edu, 40M Python-Edu, 40M FineMath). Sources: FineWeb-Edu, Python-Edu, FineMath. Source revisions/files, boundary/EOS policy, tokenizer fingerprint, counts and SHA256 checksums are recorded in training_report.json. Documents are counted separately.

Bigram PMI: minimum count 1, clip ±2. Trigram correction: count>=2, alpha=50, clip ±2. Prefix totals include singleton contributions; only prefixes needed by retained trigrams are stored. Relative trigram weight is 0.5.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support