Skip to content
Snippets Groups Projects
Select Git revision
  • add-demo-for-evaluation
  • add-tasks
  • backend
  • improve-de-cot
  • integrate-frontend
  • llama-2-format
  • llama31
  • master default protected
  • method-cot
  • refactor-models
  • refactor-task
  • user-interaction-to-optimize-operator
  • user-interaction-without-judge
13 results
You can move around the graph by using the arrow keys.
Created with Raphaël 2.2.013Feb1211109654331Jan302927242115217Dec16111063230Nov221918151413125131Oct30272524231917161110432130Sep26252423201965329Aug28222120191614986131Jul302926254Jun28May27242221171329Apr2516108419Mar181413715Feb76123Jan191816108Fix extracting prompt after last evolution step when using human interactionuser-interactio…user-interaction-without-judgeFix not using user corrected promptFix annotation data step off by 1update result notebooksmastermasterupdate result scriptsRespect annotation limit when asked to annotate multiple prompts per generationFix attribute error, remove debug prints and log additional informationMerge branch 'master' into user-interaction-without-judgeMerge branch 'user-interaction-to-optimize-operator'Fix showing wrong model inputs in user interaction when correcting prompts without judgefix run namefix syntac of evolution-demo-source parameteruser-interactio…user-interaction-to-optimize-operatorfix syntac of evolution-demo-source parameteradd trials/hf.tomladd trials/hf.tomlupdate formatting of result tablesupdate result notebooksadded human feedback result notebookadded notebook for CoT experiment resultsImprove GA-CoT with human feedback (ga-cot-hf1)Improve DE-CoT with more human feedback (de-cot-hf2)Omit evaluation history of Prompt when printed to not pollute log filesRefactor human interactions without judge and ask to correct each stepadd CoT trial configurationupdate evaluation strategies trial configupdated evaluation strategies result notebookMerge branch 'user-interaction-to-optimize-operator'[WIP] Allow human feedback without judgeadd results for evaluation strategiesMerge branch 'master' into user-interaction-to-optimize-operatorAllow setting wandb tags via CLIallow setting --evolution-demo-source to noneuse static evolution demos by defaultremove unused ForcedImprovementBasedStoppingfix hardest-first early stoppingadd subsample evaluation strategyMerge branch 'user-interaction-after-non-improvement' into 'master'add notebook to evaluate optimization strategiesadd parameters for number of evolutions and population sizeImprove de-cot with human feedback