Commit History
3 days ago
-
Add SmolLM2-135M and bloomz-560m tokenizer.json fixtures
Verified signatureByte-identical tokenizer.json from HuggingFaceTB/SmolLM2-135M@93efa2f097d58c2a74874c7e644dbc9b0cee75a2 and bigscience/bloomz-560m@a2845d7e13dd12efae154a9f1c63fcc2e0cc4b05 (main as of 2026-11-06). Both are byte-level BPE vocabularies with no token for some bytes valid UTF-8 can contain (SmolLM2: 0x04 0x06 0x13 0x14 0x16 0x1D 0xF1 0xF2; bloomz: 0x0B). The tk-convert oracle in huggingface/tokenizers (PR #2550) checks they encode like the released crate, which drops those bytes.40be28d
-
Add ModernBERT-base tokenizer fixture (#13)
- Add ModernBERT-base tokenizer fixture (768905eb5b3f61b7624fc1542574ca405a4555b1)5f5c652
-
Add siglip-base-patch16-224 and mistral-7b-v0.1 oracle fixtures (#12)
- Add siglip-base-patch16-224 and mistral-7b-v0.1 oracle fixtures (1dd19a3425aab919fde60810ed2f8a7d832da519)a6c4e9f
-
Add sentence-transformers all-MiniLM-L6-v2 and all-mpnet-base-v2 tokenizer.json fixtures (#11)
- Add sentence-transformers all-MiniLM-L6-v2 and all-mpnet-base-v2 tokenizer.json fixtures (a6a7ab03253acf671086ca5a64ca6c78208a688c)943303b
-
Add meta-llama/Llama-3.2-1B tokenizer.json fixture (#10)
- Add meta-llama/Llama-3.2-1B tokenizer.json fixture (e7644e2336dfea2d029ce1c8e59eb39f5eac5dcd)54d012f
-
Add llama-2-7b-chat-hf tokenizer.json (canonical, array-safe merges) for pipeline oracle (#6)
- Add llama-2-7b-chat-hf tokenizer.json (canonical, array-safe merges) for pipeline oracle (9ce5a3c73e84d5ed12f005e7ea29b2c1becea371) - Track fixtures/models via LFS; drop root llama-2 blob (27fedbb36343292d518ef49e189a2984b8ecf617) - Add fixtures/models/llama-2-7b-chat-hf.json (LFS) (62cf5bb1387d62adea3f16d76770360175bce4f3)a59a428