Commit Graph
2 Commits
Author SHA1 Message Date
Piotr Wilkin 10532516c3 CI/tests: refactor argument handling, add --errors-only to emit only WARN/ERR messages 2026-09-17 18:47:59 +02:00
Thiago Padilha afd439df1f unicode : include '~' in collapsed symbol class (#26972)
The collapsed \p{S} class was missing '~', which split " ~" into
separate pre-tokens and prevented the Ġ~ BPE merge used by DeepSeek V4.
This caused re-tokenized prompts to diverge from sampled tokens and
broke KV cache reuse.

Assisted-by: Codex
2026-08-18 15:15:22 +02:00