Releases: SculptAI/GIMBench
Releases · SculptAI/GIMBench
Release list
v0.5.1
v0.5.0
v0.4.0
What's Changed
- Update vllm version from 0.14.0 to 0.18.1 by @Ki-Seki in #86
- Add throughput (tokens/s) and TTFT metrics to ppl and match evaluators by @Copilot in #87
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #84
- Add HumanEval Infilling as a code generation + unit-test eval task by @Copilot in #89
- chore: bump version to 0.4.0 in pyproject.toml and uv.lock by @Ki-Seki in #94
Full Changelog: v0.3.0...v0.4.0
v0.3.0
What's Changed
- chore: ignore eval.log.* files by @Duguce in #76
- feat: add upper bound to auto reason_budget by @Duguce in #78
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #77
- Add README documentation by @Copilot in #79
- refactor: remove
richdependency by @Copilot in #81 - Unify error handling across all evaluators by @Copilot in #82
- chore: bump version to 0.3.0 by @Ki-Seki in #83
Full Changelog: v0.2.7...v0.3.0
v0.2.7
v0.2.6
What's Changed
- build: update gimkit by @Ki-Seki in #48
- refactor: change counter tokenizer & ppl model by @Ki-Seki in #51
- fix: exclude errors when calculating averages in the match task by @Ki-Seki in #49
- fix: make answer shuffling deterministic by removing python hash by @Duguce in #56
- refactor: replace CTP with windowed normalized perplexity metrics by @Ki-Seki in #52
- refactor: move counter tokenizer argument to the evaluator section by @Ki-Seki in #58
- feat: add golden truth evaluation option and update dataset column selection by @Ki-Seki in #61
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #60
- refactor: ensure results are sorted by filename before writing to CSV by @Ki-Seki in #62
- feat: add cv-parse dataset creation scripts by @Ki-Seki in #59
- feat: add CV evaluation functionality by @Ki-Seki in #63
- feat: gimbench-cv-parse dataset final pipeline by @Ki-Seki in #64
- fix: src/gimbench/cv by @Ki-Seki in #65
- feat(mcqa): improve prompt and eval robustness by @Duguce in #69
- chore: bump version to 0.2.6 by @Ki-Seki in #70
New Contributors
Full Changelog: v0.2.5...v0.2.6
v0.2.5
What's Changed
- fix: sanitize secret arguments in execution command by @Ki-Seki in #37
- refactor: remove old scripts by @Ki-Seki in #40
- fix: improve error message for context window size in CTP evaluation by @Ki-Seki in #39
- build: update deps by @Ki-Seki in #41
- feat: add max_model_len argument for model context window by @Ki-Seki in #42
- chore: bump to v0.2.5 by @Ki-Seki in #43
Full Changelog: v0.2.4...v0.2.5
v0.2.4
What's Changed
- ci: pre-commit autoupdate by @pre-commit-ci[bot] in #22
- fix: update .gitignore to remove eval results by @Ki-Seki in #24
- refactor: uniformly use SimpleGIM by @Ki-Seki in #26
- feat: add Normalized CTP by @Ki-Seki in #32
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #28
- feat: add auto-budget option by @Ki-Seki in #33
- feat: display progress information in log by @Ki-Seki in #34
- chore: bump version to 0.2.4 by @Ki-Seki in #35
Full Changelog: v0.2.3...v0.2.4
v0.2.3
What's Changed
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #19
- fix: improve git information retrieval and update eval_env structure by @Ki-Seki in #20
- chore: bump version to 0.2.3 by @Ki-Seki in #21
New Contributors
- @pre-commit-ci[bot] made their first contribution in #19
Full Changelog: v0.2.2...v0.2.3