MyAwesomeModel

Overview

MyAwesomeModel is a BERT-based model. This repository hosts the best checkpoint selected from a training sweep (steps 100–1000), chosen by the highest evaluation performance.

• Selected checkpoint: step_1000
• Selection criterion: highest overall weighted evaluation score across 15 benchmarks
• Overall weighted score: 0.710

Evaluation Results

Detailed benchmark results for the selected checkpoint (step_1000), all scores reported to three decimal places:

Category	Benchmark	Score
Core Reasoning	Math Reasoning	0.550
Core Reasoning	Logical Reasoning	0.819
Core Reasoning	Common Sense	0.736
Language Understanding	Reading Comprehension	0.700
Language Understanding	Question Answering	0.607
Language Understanding	Text Classification	0.828
Language Understanding	Sentiment Analysis	0.792
Generation	Code Generation	0.650
Generation	Creative Writing	0.610
Generation	Dialogue Generation	0.644
Generation	Summarization	0.767
Specialized	Translation	0.804
Specialized	Knowledge Retrieval	0.676
Specialized	Instruction Following	0.758
Specialized	Safety Evaluation	0.739

Overall Performance

The overall score is a weighted average with emphasis on reasoning tasks (math/logical reasoning weight 1.2, code generation, question answering, instruction following, and safety evaluation weight 1.1, others 1.0, text classification/sentiment analysis/creative writing weight 0.9).

• Overall weighted score: 0.710

Usage

from transformers import AutoModel

model = AutoModel.from_pretrained("afdadAF/MyAwesomeModel")

License

MIT