Release v0.5.3

Release 2026-07-30

Batched changes:

- Document Inkling-Small alongside Inkling

Release v0.5.2

Release v0.5.1

Release v0.5.0

- BPB (bits-per-byte) metric alongside NLL for tokenizer-independent comparison
- Modal inference integration
- Audio recipes (LibriSpeech ASR, EkaCare medical ASR) + tml_v0 renderer
- Inkling reasoning-effort controls and model-keyed rollout defaults
- RL rollout: configurable runner, parse-error policy, termination reward policy
- JWKS-based token exchange (get_access_token.py)

Load Kimi K2 tokenizers directly instead of relying on AutoTokenizer (t…

…hinking-machines-lab#785)

Add Nemotron Ultra renderer (thinking-machines-lab#776)

Fix CI: bump Kimi K2.6 revision to the `use-fast-tokenizer` commit (t…

…hinking-machines-lab#700)

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Align Kimi K2.6 revision pin (thinking-machines-lab#697)

Co-authored-by: Cursor <cursoragent@cursor.com>

Update changelog for v0.3.0 (thinking-machines-lab#645)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>