Skip to content

v0.14.0
Compare
Choose a tag to compare
@ErikBjare ErikBjare released this 21 Aug 09:21
· 512 commits to master since this release
v0.14.0
This tag was signed with the committer’s verified signature.
ErikBjare Erik Bjäreholt
GPG key ID: 73349C58108227B5
Learn about vigilant mode.
d4c55b6
This commit was signed with the committer’s verified signature.
ErikBjare Erik Bjäreholt
GPG key ID: 73349C58108227B5
Learn about vigilant mode.

These are the release notes for gptme version v0.14.0.

Contributors

Thanks to everyone who contributed to this release:

@ErikBjare

Changelog

Changes since v0.13.2:

📦 gptme

✨ Features (2)

  • feat: anthropic prompt caching beta (bb3d9a5)
  • feat: started working on vision (7b0c5b0)

🐛 Fixes (25)

Click to expand

  • fix: improved test flakiness (0758c4d)
  • fix: check if gitignore exists (d664f4b)
  • fix: moved vue@create example into bash tool (dont try to do it interactively) (5ed36eb)
  • fix: switch to ```ipython syntax for executing with python tool, to differentiate from code samples (see #67) (42ee56a)
  • fix: changed save format to be more explicit (68077b6)
  • fix: removed spammy warning (90aa4a6)
  • fix: removed use of NotRequired for TypedDict (not in Python 3.10) (cfce130)
  • fix: use anthropic api max_retries instead of custom retry_anthropic decorator (008f58c)
  • fix: added retry_anthropic decorator for rate limits (d37d7fb)
  • fix: fixed to evals, capture eval output on timeout/terminate (cd0862a)
  • fix: fixed spammy prints (855a46b)
  • fix: clarified return format for subagent (e5e2a9a)
  • fix: fixed bug in transform_examples_to_chat_directives (2997aa4)
  • fix: print logs dir with --version command (1ca6127)
  • fix: enabled stricter linting and fixed lints (apparently needed given a25aa7d) (bf67b32)
  • fix: fixed nasty bug with mutable argument default (a25aa7d)
  • fix: switch recommended openai model to gpt-4o (b3582ac)
  • fix: moved len_tokens and msgs2dicts from util.py to message.py (94bade3)
  • fix: support nested codeblocks, rewrote/refactored codeblock parsing/management (3e291a4)
  • fix: fixed incorrect storage format for Message.to_dict (07f1cbb)
  • fix: added preliminary nested codeblock support (885e544)
  • fix: added openai vision support (6bbec93)
  • fix: completed basic vision support (f184607)
  • fix: correct extremely short 'Thinking...' message, now wait until first character (b25e576)
  • fix: typing for subagent tool, added --check-untypes-defs to mypy (d94a71e)

🔨 Misc (20)

Click to expand

  • chore: bumped version to v0.14.0 (d4c55b6)
  • refactor: renamed terminal tool to tmux (79355bc)
  • tests: increase max tokens for full system prompt significantly (c785cfe)
  • tests: made test_subprocess less flaky (3423474)
  • tests: added difficult but reasonably simple integration test example (create vite project, build simple app) (57b6feb)
  • docs(README): minor improvements (c43cc8a)
  • docs(README): misc minor improvements (c705c80)
  • tests: removed uninteresting examples from test-integration.sh (18c1afa)
  • docs: added links to top of README (b15d686)
  • docs(README): added mention that vision works, 'GPTMe' -> 'gptme' for title, mention OpenRouter support (4e39f63)
  • tests: run both eval tests (cli and direct call) for better coverage somehow (859efbe)
  • tests: fixed tests failing due to new file (706f13c)
  • refactor: refactored provider-specific code into new files llm_openai.py and llm_anthropic.py (eec8215)
  • tests: fix coverage for eval tests (multiprocessing workaround) (9ef1ec4)
  • tests: added test_eval_cli and cleaned up eval code (c0e5c87)
  • tests: basic test for evals (770ed2c)
  • tests: fix --version test (4ca5078)
  • tests: remove ambiguity in subagent fib test prompt (fe063ec)
  • tests: added test for vision (a740194)
  • refactor: moved eval code into gptme.eval, added gptme-eval entrypoint, fixed typing (8a1bb09)

(excluded 14 less relevant commits)

Full Changelog: v0.13.2...v0.14.0

Contributors

ErikBjare
Assets 2
Loading