VAPOR benchmark data, pretrained reward models, and ESM-2 logits cache for protein sequence optimization

This dataset provides the benchmark data and pretrained assets used by
VAPOR, a Python toolkit for Pareto-leading protein sequence optimization
without invoking a protein language model during search.

The deposit contains three archives:

1. vapor_benchmark_data.tar.gz:
   Benchmark DMS datasets used for protein sequence optimization
   experiments, including core benchmarks and prepared ProteinGym
   landscapes.

2. vapor_pretrained_models.tar.gz:
   Pretrained reward model ensemble checkpoints used by VAPOR.

3. vapor_esm2_logits_cache.tar.gz:
   Cached ESM-2 per-position token logits used for offline PLM
   distillation. These caches avoid runtime PLM calls during sequence
   optimization.

These files are intended to be extracted into the VAPOR repository root
or release package directory before running the command-line interface.
The source code is available separately as part of the VAPOR software
package.