diff --git a/tools/submission/submission_checker.py b/tools/submission/submission_checker.py index d100d3ea0..68e446117 100755 --- a/tools/submission/submission_checker.py +++ b/tools/submission/submission_checker.py @@ -735,19 +735,19 @@ ACC_PATTERN = { "acc": r"^(?:\{\"accuracy|accuracy)[\": ]*=?\s*([\d\.]+).*", "AUC": r"^AUC=([\d\.]+).*", - "mAP": r".*'(?:mAP|Total)':\s*([\d\.]+)", + "mAP": r".*(?:mAP=|Total:)\s*([\d.]+)", "bleu": r"^BLEU\:\s*([\d\.]+).*", "F1": r"^{[\"\']exact_match[\"\']\:\s*[\d\.]+,\s*[\"\']f1[\"\']\:\s*([\d\.]+)}", "WER": r"Word Error Rate\:.*, accuracy=([0-9\.]+)%", "DICE": r"Accuracy\:\s*mean\s*=\s*([\d\.]+).*", - "ROUGE1": r".*'rouge1':\s([\d.]+).*", - "ROUGE2": r".*'rouge2':\s([\d.]+).*", - "ROUGEL": r".*'rougeL':\s([\d.]+).*", + "ROUGE1": r".*'rouge1':\s+'?([\d.]+)'?.*", + "ROUGE2": r".*'rouge2':\s+'?([\d.]+)'?.*", + "ROUGEL": r".*'rougeL':\s+'?([\d.]+)'?.*", "ROUGELSUM": r".*'rougeLsum':\s([\d.]+).*", "GEN_LEN": r".*'gen_len':\s([\d.]+).*", "TOKENS_PER_SAMPLE": r".*'tokens_per_sample':\s([\d.]+).*", - "CLIP_SCORE": r".*'CLIP_SCORE':\s([\d.]+).*", - "FID_SCORE": r".*'FID_SCORE':\s([\d.]+).*", + "CLIP_SCORE": r".*'CLIP_SCORE':\s.([\d.]+).*", + "FID_SCORE": r".*'FID_SCORE':\s.([\d.]+).*", "gsm8k_accuracy": r".*'gsm8k':\s([\d.]+).*", "mbxp_accuracy": r".*'mbxp':\s([\d.]+).*", "exact_match": r".*'exact_match':\s([\d.]+).*"