Add HAKARI-Bench results for hotchpotch/japanese-splade-v2

Summary

Core nDCG@10

Reproducibility

Command

# Reconstructed from result metadata; exact shell history was not recorded in the result files.
uv run hakari-bench evaluate sparse \
  --model hotchpotch/japanese-splade-v2 \
  --all \
  --dtype bf16 \
  --device cuda:0 \
  --batch-size 32 \
  --attn-implementation sdpa \
  --model-max-seq-length 512

Submitter Notes

• Sparse evaluation for hotchpotch/japanese-splade-v2.
• Batch size is 32 for 550 files and 8 for 1 file, indicating one task was rerun or resumed with a smaller batch.
• Submitted files cover 551 .json.xz task results.
• No DuckDB, cache files, HTML reports, YAML model config, or scratch artifacts are included.
• The Core summary above was computed from the submitted result files.

Checklist

• Result files are committed under hakari-results/hotchpotch__japanese-splade-v2/.
• Result files are compressed .json.xz.
• Result JSON records model revision, dataset revision, runtime configuration, and package versions.
• Core nDCG@10 above was generated from the submitted result files.
• Non-default runtime choices are documented above.