Buckets:
6.24 TB
2,775 files
Updated about 1 month ago
Ctrl+K
| Name | Size | Uploaded | Xet hash |
|---|---|---|---|
| README.md | 416 Bytes xet | e6f05ca3 | |
| embedding_bench.py | 1.45 kB xet | 5596d947 | |
| report.pdf | 6.02 kB xet | 68a43437 | |
| results.json | 1.25 kB xet | 2f7f392a | |
| run.py | 1.67 kB xet | d32fe90b | |
| wordsim353.csv | 7.4 kB xet | 32964209 |
Benchmarks
results.json holds every number on the model card; report.pdf is the full report.
The evaluation archives are held out and not distributed.
run.py is the harness that produced the numbers. You can point it at your own archive
(a folder with a tokens.u32 stream and a bank_valid.jsonl question bank in the same
format as results.json describes) to run the same evaluation on your own data.
- Total size
- 6.24 TB
- Files
- 2,775
- Last updated
- Sep 8
- Pre-warmed CDN
- US EU US EU