E2LM: Early Training Evaluation of Language Models - building new scientific benchmarks for Small Language Models with the community