Hugging Face's logo

Datasets:
hakari-bench
/
results

Dataset card Files
xet
 Community
50

Add benchmark results for perplexity-ai/pplx-embed-v2-late-0.6b

#49
by hotchpotch - opened
base: refs/heads/main
←
from: refs/pr/49
Discussion Files changed
+1689
-0
Add benchmark results for perplexity-ai/pplx-embed-v2-late-0.6bfc952963
hotchpotch
 HAKARI-Bench org

Add HAKARI-Bench results for perplexity-ai/pplx-embed-v2-late-0.6b

Summary

Field Value
Model perplexity-ai/pplx-embed-v2-late-0.6b
Result directory perplexity-ai__pplx-embed-v2-late-0.6b
Target path hakari-results/perplexity-ai__pplx-embed-v2-late-0.6b
Result files 563 total, 563 .json.xz
Evaluation method late-interaction
Overall nDCG@10 0.6375
Overall score units 381 grouped units from 550 raw task results

DuckDB Nano-set Comparison

Computed from DuckDB task_results with the same Overall grouping as this PR body. Quantized and rescore variants are excluded; truncate variants are considered, and each model column uses that model's best Overall variant.

Overall component perplexity-ai/pplx-embed-v2-late-0.6b Qwen/Qwen3-Embedding-0.6B (1024 dims) jinaai/jina-embeddings-v5-text-small (1024 dims) BAAI/bge-m3 (1024 dims) intfloat/multilingual-e5-small (384 dims) bm25
Overall 0.6375 0.5944 0.6296 0.5820 0.5149 0.4806
NanoMMTEB-v2 0.5522 0.5581 0.5590 0.4846 0.4455 0.4550
NanoRTEB 0.6137 0.6713 0.7005 0.5365 0.4711 0.3553
MNanoBEIR 0.6058 0.5509 0.6077 0.5575 0.5117 0.4646
NanoBIRCO 0.3809 0.3070 0.3526 0.2617 0.1613 0.2693
NanoMLDR 0.6356 0.6239 0.5384 0.6621 0.3920 0.7396
NanoLongEmbed 0.7149 0.7232 0.6680 0.6527 0.5014 0.8217
NanoDAPFAM 0.3157 0.3018 0.3179 0.2406 0.2380 0.2400
NanoCoIR 0.6643 0.8601 0.8777 0.6924 0.6915 0.5436
NanoIFIR 0.4227 0.3364 0.3893 0.2391 0.2152 0.2761
NanoLaw 0.7722 0.6075 0.6370 0.5597 0.4790 0.6854
NanoMedical 0.5953 0.5694 0.5803 0.5371 0.5055 0.4145
NanoRARb 0.2280 0.2689 0.2889 0.2343 0.2240 0.1359
NanoBRIGHT 0.3812 0.3885 0.4284 0.2941 0.1758 0.2790
NanoCodeRAG 0.8781 0.8712 0.9139 0.7155 0.7464 0.5823
NanoChemTEB 0.8328 0.8035 0.7980 0.7777 0.8081 0.7012
NanoR2MED 0.3470 0.3180 0.3630 0.2088 0.1099 0.2094
NanoBuiltBench 0.5503 0.5129 0.5277 0.4248 0.4291 0.3958
NanoCMTEB 0.7948 0.7982 0.8052 0.7591 0.6999 0.6003
NanoIndicQA 0.8311 0.6413 0.7056 0.7586 0.7009 0.5653
NanoMuPLeR 0.9574 0.7122 0.8388 0.8912 0.7837 0.7994
NanoMTEB-v2 0.6360 0.6372 0.6450 0.5726 0.5348 0.5028
NanoMTEB-Dutch 0.6431 0.5686 0.6213 0.5863 0.5287 0.4673
NanoMTEB-French 0.6871 0.5771 0.6377 0.5527 0.4702 0.4261
NanoMTEB-German 0.7129 0.6298 0.6536 0.6189 0.5711 0.5522
NanoJMTEB-v2 0.8278 0.7732 0.8008 0.7906 0.7165 0.7465
NanoMTEB-Korean 0.8663 0.7792 0.8246 0.8183 0.7668 0.6743
NanoFaMTEB-v2 0.6888 0.6338 0.6882 0.6652 0.6135 0.5651
NanoMTEB-Polish 0.5571 0.4738 0.5316 0.4999 0.4365 0.3424
NanoMTEB-BR 0.7359 0.6247 0.6595 0.6540 0.5224 0.5178
NanoSSRB 0.3874 0.3464 0.4340 0.2665 0.2511 0.2821
NanoRuMTEB 0.9184 0.8622 0.9121 0.9169 0.8643 0.7089
NanoMTEB-Scandinavian 0.7702 0.6981 0.7596 0.7740 0.7029 0.6091
NanoMTEB-Spanish 0.6683 0.5662 0.6292 0.5624 0.4848 0.3679
NanoMTEB-Thai 0.7853 0.7455 0.7670 0.7672 0.7107 0.5216
NanoVNMTEB 0.6192 0.5717 0.6066 0.5616 0.5197 0.4571
NanoMTEB-Misc 0.8013 0.7629 0.8011 0.7766 0.6423 0.4939
NanoMIRACL 0.8303 0.7879 0.8351 0.8475 0.7871 0.5715

Overall nDCG@10

Overall component nDCG@10 Score units Raw task results
NanoMMTEB-v2 0.5522 18 18
NanoRTEB 0.6137 14 14
MNanoBEIR 0.6058 13 182
NanoBIRCO 0.3809 5 5
NanoMLDR 0.6356 13 13
NanoLongEmbed 0.7149 6 6
NanoDAPFAM 0.3157 12 12
NanoCoIR 0.6643 10 10
NanoIFIR 0.4227 4 4
NanoLaw 0.7722 4 4
NanoMedical 0.5953 7 7
NanoRARb 0.2280 14 14
NanoBRIGHT 0.3812 20 20
NanoCodeRAG 0.8781 4 4
NanoChemTEB 0.8328 3 3
NanoR2MED 0.3470 8 8
NanoBuiltBench 0.5503 2 2
NanoCMTEB 0.7948 8 8
NanoIndicQA 0.8311 11 11
NanoMuPLeR 0.9574 14 14
NanoMTEB-v2 0.6360 10 10
NanoMTEB-Dutch 0.6431 27 27
NanoMTEB-French 0.6871 8 8
NanoMTEB-German 0.7129 5 5
NanoJMTEB-v2 0.8278 11 11
NanoMTEB-Korean 0.8663 5 5
NanoFaMTEB-v2 0.6888 17 17
NanoMTEB-Polish 0.5571 14 14
NanoMTEB-BR 0.7359 6 6
NanoSSRB 0.3874 6 6
NanoRuMTEB 0.9184 3 3
NanoMTEB-Scandinavian 0.7702 7 7
NanoMTEB-Spanish 0.6683 7 7
NanoMTEB-Thai 0.7853 9 9
NanoVNMTEB 0.6192 26 26
NanoMTEB-Misc 0.8013 12 12
NanoMIRACL 0.8303 18 18

Reproducibility

Field Value
Model source perplexity-ai/pplx-embed-v2-late-0.6b
Model revision 7bba8e7cf2075dec747fa8557f41eb90e70361ed
Dataset revision(s) 00541a0fce4048057fb7ddec30d37155a5c23d95, 01736efbaa96f020c2a4d996efdacc18071e2fcb, 017849a95097eea984680cbab35972f8d3812376, 0f3a6f43b8a26a9b8c8d5f31b09bd60dc4cd572d, 1726763179e1e114ad9ffcdc7262923471e8ecc8, ... (50 total)
Evaluated at UTC 2026-11-04T06:16:56.532512+00:00 to 2026-11-04T10:37:18.347318+00:00
Generated at UTC 2026-11-04T06:16:56.532539+00:00 to 2026-11-04T10:37:18.347342+00:00
dtype fp32
device cuda:0
batch size 2, 8
attention implementation sdpa
trust remote code False
max sequence length 4096
candidate ranking reranking_hybrid
rerank top-k not recorded
query prompt name not recorded
document prompt name not recorded
Python 3.12.15 (main, Oct 21 2026, 17:47:06) [Clang 23.1.3 ]
Platform Linux-6.8.0-147-generic-x86_64-with-glibc2.39
torch 2.9.1
transformers 5.22.1
sentence-transformers 6.2.0
datasets 5.1.0
CUDA available=True, version=12.8
CUDA devices 0: NVIDIA GeForce RTX 5090

Command 

# Python argv manifest (one call per pinned dataset/task; two GPU workers):
command=['evaluate','late-interaction','--model',plan['model'],'--model-revision',plan['model_revision'],
  '--model-loader','pplx_backend:load_model',
  '--dataset',item['dataset'],'--split',item['task'],'--dataset-revision',plan['dataset_revisions'][item['dataset']],
  '--dtype','fp32','--attn-implementation','sdpa','--device','cuda:0','--batch-size',str(batch),
  '--late-interaction-exact-doc-batch-size','16','--late-interaction-exact-query-batch-size','2',
  '--results-dir',str(out)]

Submitter Notes

  • Native ST6 MultiVectorEncoder role methods via the adapter in code PR; FP32/SDPA, query/document limits 1024/4096, no query expansion. Optimized FLA/causal-conv1d installed. MaxSim preflight matches native within 2.4e-7. Batch 8 for short tasks, 2 for long tasks.
  • Both GPUs used a shared small-first task queue; all tasks completed without errors. Context was not shortened. Model reused and warmed in each worker.
  • Standard full scope: 563 task files, including 13 overlapping NanoBEIR-en results retained for completeness. All 50 dataset revisions pinned and recorded in per-task metadata.

Checklist

  • Result files are committed under hakari-results/perplexity-ai__pplx-embed-v2-late-0.6b/.
  • Result files are compressed .json.xz; no caches, DuckDB files, HTML reports, or local scratch artifacts are included.
  • The result JSON records model revision, dataset revision, runtime configuration, and package versions.
  • Overall nDCG@10 above was generated from the submitted result files.
  • Any non-default prompt, sequence length, attention implementation, candidate ranking, or reranker setting is documented above.

Code, model cards and reusable loader: https://github.com/hakari-bench/hakari-bench/pull/19
hotchpotch changed pull request status to merged

Sign up or log in to comment