Skip to content

v0.11.0

Latest

Choose a tag to compare

@github-actions github-actions released this 17 Nov 12:31
· 1 commit to main since this release
5961eef

What's Changed

  • fix(server): keep tool requests off the batched path by @magicnight in #157
  • feat(server): OpenAI Responses API for Codex by @magicnight in #158
  • feat(server): Claude Code on /v1/messages, and count_tokens by @magicnight in #159
  • feat(cache): reuse agent-turn history on hybrid and sliding-window models by @magicnight in #160
  • feat(server): carry replayed reasoning into the chat template by @magicnight in #161
  • Make Xcode 27 and Apple's Foundation Models client work against macMLX by @magicnight in #162
  • Match the VLM type list to mlx-swift-lm's registry by @magicnight in #163
  • Design images for Gemma 4 and Qwen3.6 on the full serving path by @magicnight in #164
  • feat(engine): serve a VLM on the text generation path behind a seam by @magicnight in #165
  • feat(engine): reuse the prompt cache up to an image request's first block by @magicnight in #166
  • feat(models): images on Gemma 4 E4B on the full serving path by @magicnight in #167
  • test(engine): run the full-path gates on Qwen3.6 and record why it stays text by @magicnight in #168
  • fix(models): keep a Gemma 4 config without use_double_wide_mlp on text by @magicnight in #169
  • test(engine): run the other Gemma 4 sizes; load Gemma 4 MoE as text by @magicnight in #170
  • fix(engine): prefill the text before an image in chunks by @magicnight in #171
  • test(engine): build tiny random models from their own random state by @magicnight in #172
  • test(models): draw parity inputs from a random state of their own by @magicnight in #173
  • feat(models): images on dense Qwen3.5/3.6 by @magicnight in #174
  • fix(engine): pass JSON null to the chat template as nil by @magicnight in #176
  • chore(release): 0.11.0 by @magicnight in #177

Full Changelog: v0.10.0...v0.11.0