213DSA12EW/bert-base-uncased-step1000-best

BERT Model Checkpoint - Step 1000

This model checkpoint (step_1000) was selected as the best performing model based on evaluation accuracy across all benchmark categories.

Model Details

• Model Type: BERT (BertModel)
• Checkpoint Step: 1000
• Overall Weighted Score: 0.806

Evaluation Results

Detailed evaluation scores across all 15 benchmark categories (scores rounded to 3 decimal places):

Benchmark Category	Score	Weight	Weighted Score
code_generation	0.769	1.1	0.846
common_sense	0.794	1.0	0.794
creative_writing	0.729	0.9	0.656
dialogue_generation	0.745	1.0	0.745
instruction_following	0.818	1.1	0.9
knowledge_retrieval	0.834	1.0	0.834
logical_reasoning	0.81	1.2	0.972
math_reasoning	0.81	1.2	0.972
question_answering	0.851	1.1	0.936
reading_comprehension	0.826	1.0	0.826
safety_evaluation	0.802	1.1	0.882
sentiment_analysis	0.891	0.9	0.802
summarization	0.761	1.0	0.761
text_classification	0.875	0.9	0.787
translation	0.778	1.0	0.778

Summary Statistics

• Number of Benchmarks: 15
• Highest Score: 0.891 (sentiment_analysis)
• Lowest Score: 0.729 (creative_writing)
• Average Score: 0.806
• Overall Weighted Score: 0.806

Checkpoint Selection

This checkpoint was selected as the best model because it achieved the highest evaluation accuracy among all checkpoints evaluated (steps 100 through 1000).