Jun 30 12:57 preliminary_mul_200k: replace with v200064 (vocab 200064, 128-aligned; IDs 0-199999 identical to prior 200000 build) + same special tokens + post-processor Jun 25 10:13 preliminary_mul_200k: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_mul_200k: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_mul: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_mul: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_euh: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_euh: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_enh: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> 10:13 preliminary_enh: rename PII tokens to spreadsheet order <iban-pii>/<email-pii>/<ip-pii> Jun 22 11:09 preliminary_mul_200k: replace eusino_v2c with frde_kr120_gm130 (French at Apertus parity, German/Spanish below; Mandarin>Korean ordering kept) Jun 21 20:23 preliminary_mul_200k: PII special-token rename (tokenizer_config) 20:23 preliminary_mul_200k: rename reserve specials 24/25/26 to <pii-iban>/<pii-email>/<pii-ip> 20:23 preliminary_mul: PII special-token rename (tokenizer_config) 20:23 preliminary_mul: rename reserve specials 24/25/26 to <pii-iban>/<pii-email>/<pii-ip> 20:23 preliminary_euh: PII special-token rename (tokenizer_config) 20:23 preliminary_euh: rename reserve specials 24/25/26 to <pii-iban>/<pii-email>/<pii-ip> 20:23 preliminary_enh: PII special-token rename (tokenizer_config) 20:23 preliminary_enh: rename reserve specials 24/25/26 to <pii-iban>/<pii-email>/<pii-ip> 15:39 preliminary_mul: replace with reparam variant (consv2_reparam, v131072, full vocab) + post-processor 09:17 preliminary_mul: add BOS/EOS post-processor (single <s> $A </s>, pair seg-B type_id 1) — now consistent with the other three Jun 19 20:37 preliminary_euh: replace with frde2 (131k, Fr/De data-boosted, EU6 -0.2; Chinese the cost) + BOS/EOS post-processor 12:29 add preliminary_enh (engfull_eu3, 131k English-preserving) + preliminary_mul_200k (eusino_v2c, 200k), both with BOS/EOS post-processor Jun 18 14:04 rename preliminary_enh -> preliminary_euh 10:25 preliminary_enh: pair post-processor type_id -> segment B = 1 (Apertus convention); single unchanged 10:09 preliminary_enh: replace with eudata6_gm90k (fixed plus2 repcap8 regex + BOS/EOS post-processor); vocab now exactly 131072 Jun 16 13:30 Fix eos_token to </s> (id 2) in preliminary_enh + preliminary_mul configs (was wrongly <|assistant_end|>); special_tokens_map regenerated to match base Apertus-70B-2509 12:51 Add tokenizer_config.json + special_tokens_map.json for preliminary_enh and preliminary_mul (eos=<|assistant_end|>, add_bos_token=false interim pending post-processor decision) Jun 8 19:09 Add preliminary_mul — consv2 baseline + eng5g, no tailcuts, vocab 131k sp124 (fixed regex) 19:09 Replace preliminary_enh with A8 v131k sp124 (fixed regex + 124 specials at IDs 0-123) 13:29 Add preliminary enhanced tokenizer (A8 repcap8 v131k sp120) 13:29 initial commit 
