ASD123SAD21/MyAwesomeModel-TestRepository

MyAwesomeModel

Best Checkpoint

The best checkpoint selected is step_1000, which achieves the highest overall weighted eval_accuracy across all 15 benchmark categories.

Overall weighted eval_accuracy (step_1000): 0.710

Evaluation Results — Best Checkpoint (step_1000)

All scores below are computed from the benchmark scoring functions and rounded to three decimal places.

#	Benchmark Category	eval_accuracy
1	math_reasoning	0.550
2	code_generation	0.650
3	text_classification	0.828
4	sentiment_analysis	0.792
5	question_answering	0.607
6	logical_reasoning	0.819
7	common_sense	0.736
8	reading_comprehension	0.700
9	dialogue_generation	0.644
10	summarization	0.767
11	translation	0.804
12	knowledge_retrieval	0.676
13	creative_writing	0.610
14	instruction_following	0.758
15	safety_evaluation	0.739

Weighting Used for Overall Score

The overall weighted eval_accuracy combines the per-benchmark scores with the following weights (slightly emphasizing reasoning tasks):

Benchmark	Weight
math_reasoning	1.2
logical_reasoning	1.2
code_generation	1.1
question_answering	1.1
instruction_following	1.1
safety_evaluation	1.1
reading_comprehension	1.0
common_sense	1.0
dialogue_generation	1.0
summarization	1.0
translation	1.0
knowledge_retrieval	1.0
text_classification	0.9
sentiment_analysis	0.9
creative_writing	0.9

Per-Category eval_accuracy Trend Across Checkpoints

For reference, the overall weighted eval_accuracy for every checkpoint found in the workspace:

Checkpoint	Overall eval_accuracy
step_100	0.480
step_200	0.535
step_300	0.576
step_400	0.608
step_500	0.635
step_600	0.656
step_700	0.674
step_800	0.689
step_900	0.700
step_1000	0.710

Architecture

• model_type: bert
• architectures: ["BertModel"]