From fc3a43ef41a2580149ad5d1cd5dd5bad9b07c0a9 Mon Sep 17 00:00:00 2001 From: Paul Gauthier Date: Mon, 6 May 2024 11:25:14 -0700 Subject: [PATCH] completed moving refac to yml --- _data/refactor_leaderboard.yml | 62 +++++++++++++++++++++++++++++++++- 1 file changed, 61 insertions(+), 1 deletion(-) diff --git a/_data/refactor_leaderboard.yml b/_data/refactor_leaderboard.yml index 01ae9dbf1..f259557e8 100644 --- a/_data/refactor_leaderboard.yml +++ b/_data/refactor_leaderboard.yml @@ -33,9 +33,69 @@ indentation_errors: 3 exhausted_context_windows: 0 test_timeouts: 0 - command: aider --model openrouter/anthropic/claude-3-opus + command: aider --opus date: 2024-05-04 versions: 0.31.2-dev seconds_per_case: 67.8 total_cost: 27.9176 +- dirname: 2024-04-09-21-49-54--refac-gpt-4-turbo-2024-04-09 + test_cases: 88 + model: gpt-4-turbo-2024-04-09 + edit_format: udiff + commit_hash: b75fdb9 + pass_rate_1: 34.1 + percent_cases_well_formed: 30.7 + error_outputs: 183 + num_malformed_responses: 61 + user_asks: 0 + lazy_comments: 1 + syntax_errors: 3 + indentation_errors: 15 + exhausted_context_windows: 0 + test_timeouts: 0 + command: aider --gpt-4-turbo + date: 2024-04-09 + versions: 0.27.1-dev + seconds_per_case: 42.4 + total_cost: 19.6556 +- dirname: 2024-01-25-22-07-21--jan-gpt-4-0125-preview-udiff + test_cases: 89 + model: gpt-4-0125-preview + edit_format: udiff + commit_hash: 0fbd702 + pass_rate_1: 43.8 + percent_cases_well_formed: 74.2 + error_outputs: 51 + num_malformed_responses: 23 + user_asks: 0 + lazy_comments: 1 + syntax_errors: 4 + indentation_errors: 11 + exhausted_context_windows: 1 + test_timeouts: 0 + command: aider --model gpt-4-0125-preview + date: 2024-01-25 + versions: 0.22.1-dev + seconds_per_case: 70.4 + total_cost: 43.3437 +- dirname: 2024-01-25-21-27-47--jan-gpt-4-1106-preview-udiff + test_cases: 89 + model: gpt-4-1106-preview + edit_format: udiff + commit_hash: a75e7c8 + pass_rate_1: 57.3 + percent_cases_well_formed: 31.5 + error_outputs: 127 + num_malformed_responses: 61 + user_asks: 0 + lazy_comments: 4 + syntax_errors: 1 + indentation_errors: 15 + exhausted_context_windows: 1 + test_timeouts: 0 + command: aider + date: 2024-01-25 + versions: 0.22.1-dev + seconds_per_case: 181.9 + total_cost: 18.6347 \ No newline at end of file