Skip to content

v0.0.7

Latest
Latest

Choose a tag to compare

View all tags
@RobertTLange RobertTLange released this 02 Jun 16:44
· 104 commits to main since this release
v0.0.7
c4568ad

PyPI: https://pypi.org/project/shinka-evolve/0.0.7/

Added

  • Added Gemini 3.5 Flash pricing to the Google LLM pricing catalog.
  • Added Claude Opus 4.8 pricing entries for the Anthropic API and Amazon Bedrock (us.anthropic.claude-opus-4-8) in the LLM pricing catalog.
  • Added Wolfram Language as a first-class evolution target: registry entries (wolfram, wl, wls, mathematica), code-fence and EVOLVE-BLOCK marker support, and end-to-end coverage in apply_diff / apply_full.
  • Added EVOLVE-BLOCK marker validation in shinka.edit.marker_validation, catching the LLM failure mode where a block-comment-language marker is emitted without its closing delimiter (Wolfram, Markdown), which would silently trap the candidate body inside a comment.
  • Added the examples/wolfram_gcd_sum task: deoptimized seed for S(N) = sum_{i,j in 1..N} GCD(i,j). The evaluator calibrates the baseline on every run by timing the seed through the same RepeatedTiming harness as the candidate, and the Wolfram-side timeout is configurable via WOLFRAM_GCD_MAX_SECONDS.
  • Added Go as a first-class evolution target with go/golang language registration, .go task detection, WebUI failure-artifact support, regression coverage, and the examples/go_collatz_steps task.

Fixed

  • Fixed the WebUI dashboard so runs that have completed only the initial generation remain visible instead of being filtered out as empty results.
  • Fixed the WebUI Ensembling tab so single-model Headless runs render visible data points and keep the top y-axis tick labels in view.
Assets 2
Loading