Hugging Face's logo

Datasets:
hakari-bench
/
results

Dataset card Files
xet
 Community
29

Add results for nvidia/Nemotron-3-Embed-1B-BF16

#26
by hotchpotch - opened
base: refs/heads/main
←
from: refs/pr/26
Discussion Files changed
+1653
-0
Add results for nvidia/Nemotron-3-Embed-1B-BF165b77819c
hotchpotch
 HAKARI-Bench org

Add HAKARI-Bench results for nvidia/Nemotron-3-Embed-1B-BF16

Summary

Field Value
Model nvidia/Nemotron-3-Embed-1B-BF16
Result directory nvidia__Nemotron-3-Embed-1B-BF16
Target path hakari-results/nvidia__Nemotron-3-Embed-1B-BF16
Result files 551 total, 551 .json.xz
Evaluation method dense
Overall nDCG@10 0.6190
Overall score units 369 grouped units from 538 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 nvidia/Nemotron-3-Embed-1B-BF16 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 0.6190 0.5979 0.6323 0.5859 0.5190 0.4832
NanoMMTEB-v2 0.6235 0.5581 0.5590 0.4846 0.4455 0.4550
NanoRTEB 0.7492 0.6713 0.7005 0.5365 0.4711 0.3553
MNanoBEIR 0.5736 0.5509 0.6077 0.5575 0.5117 0.4646
NanoBIRCO 0.3447 0.3070 0.3526 0.2617 0.1613 0.2693
NanoMLDR 0.6053 0.6239 0.5384 0.6621 0.3920 0.7396
NanoLongEmbed 0.7696 0.7232 0.6680 0.6527 0.5014 0.8217
NanoDAPFAM 0.3108 0.3018 0.3179 0.2406 0.2380 0.2400
NanoCoIR 0.8873 0.8601 0.8777 0.6924 0.6915 0.5436
NanoIFIR 0.3694 0.3364 0.3893 0.2391 0.2152 0.2761
NanoLaw 0.6892 0.6075 0.6370 0.5597 0.4790 0.6854
NanoMedical 0.5383 0.5694 0.5803 0.5371 0.5055 0.4145
NanoRARb 0.3008 0.2689 0.2889 0.2343 0.2240 0.1359
NanoBRIGHT 0.4161 0.3885 0.4284 0.2941 0.1758 0.2790
NanoCodeRAG 0.9030 0.8712 0.9139 0.7155 0.7464 0.5823
NanoChemTEB 0.8503 0.8035 0.7980 0.7777 0.8081 0.7012
NanoR2MED 0.3697 0.3180 0.3630 0.2088 0.1099 0.2094
NanoBuiltBench 0.5285 0.5129 0.5277 0.4248 0.4291 0.3958
NanoCMTEB 0.7231 0.7982 0.8052 0.7591 0.6999 0.6003
NanoIndicQA 0.5990 0.6413 0.7056 0.7586 0.7009 0.5653
NanoMuPLeR 0.6880 0.7122 0.8388 0.8912 0.7837 0.7994
NanoMTEB-v2 0.6300 0.6372 0.6450 0.5726 0.5348 0.5028
NanoMTEB-Dutch 0.6117 0.5686 0.6213 0.5863 0.5287 0.4673
NanoMTEB-French 0.6469 0.5771 0.6377 0.5527 0.4702 0.4261
NanoMTEB-German 0.6812 0.6298 0.6536 0.6189 0.5711 0.5522
NanoJMTEB-v2 0.8143 0.7732 0.8008 0.7906 0.7165 0.7465
NanoMTEB-Korean 0.8591 0.7792 0.8246 0.8183 0.7668 0.6743
NanoFaMTEB-v2 0.6790 0.6338 0.6882 0.6652 0.6135 0.5651
NanoMTEB-Polish 0.3455 0.4738 0.5316 0.4999 0.4365 0.3424
NanoRuMTEB 0.9300 0.8622 0.9121 0.9169 0.8643 0.7089
NanoMTEB-Scandinavian 0.6920 0.6981 0.7596 0.7740 0.7029 0.6091
NanoMTEB-Spanish 0.6115 0.5662 0.6292 0.5624 0.4848 0.3679
NanoMTEB-Thai 0.7529 0.7455 0.7670 0.7672 0.7107 0.5216
NanoVNMTEB 0.5861 0.5717 0.6066 0.5616 0.5197 0.4571
NanoMTEB-Misc 0.7917 0.7629 0.8011 0.7766 0.6423 0.4939
NanoMIRACL 0.8473 0.7879 0.8351 0.8475 0.7871 0.5715

Overall nDCG@10

Overall component nDCG@10 Score units Raw task results
NanoMMTEB-v2 0.6235 18 18
NanoRTEB 0.7492 14 14
MNanoBEIR 0.5736 13 182
NanoBIRCO 0.3447 5 5
NanoMLDR 0.6053 13 13
NanoLongEmbed 0.7696 6 6
NanoDAPFAM 0.3108 12 12
NanoCoIR 0.8873 10 10
NanoIFIR 0.3694 4 4
NanoLaw 0.6892 4 4
NanoMedical 0.5383 7 7
NanoRARb 0.3008 14 14
NanoBRIGHT 0.4161 20 20
NanoCodeRAG 0.9030 4 4
NanoChemTEB 0.8503 3 3
NanoR2MED 0.3697 8 8
NanoBuiltBench 0.5285 2 2
NanoCMTEB 0.7231 8 8
NanoIndicQA 0.5990 11 11
NanoMuPLeR 0.6880 14 14
NanoMTEB-v2 0.6300 10 10
NanoMTEB-Dutch 0.6117 27 27
NanoMTEB-French 0.6469 8 8
NanoMTEB-German 0.6812 5 5
NanoJMTEB-v2 0.8143 11 11
NanoMTEB-Korean 0.8591 5 5
NanoFaMTEB-v2 0.6790 17 17
NanoMTEB-Polish 0.3455 14 14
NanoRuMTEB 0.9300 3 3
NanoMTEB-Scandinavian 0.6920 7 7
NanoMTEB-Spanish 0.6115 7 7
NanoMTEB-Thai 0.7529 9 9
NanoVNMTEB 0.5861 26 26
NanoMTEB-Misc 0.7917 12 12
NanoMIRACL 0.8473 18 18

Reproducibility

Field Value
Model source nvidia/Nemotron-3-Embed-1B-BF16
Model revision 0677b2025cbc37daf92d7b9c7a225de8cfbc5b0b
Dataset revision(s) 01736efbaa96f020c2a4d996efdacc18071e2fcb, 017849a95097eea984680cbab35972f8d3812376, 0f3a6f43b8a26a9b8c8d5f31b09bd60dc4cd572d, 1726763179e1e114ad9ffcdc7262923471e8ecc8, 175ff423246cdbca9c3a992c4d68d312701b3f2a, ... (48 total)
Evaluated at UTC 2026-07-17T03:24:39.023959+00:00 to 2026-07-17T15:41:11.267690+00:00
Generated at UTC 2026-07-17T03:24:39.242149+00:00 to 2026-07-17T15:41:11.267709+00:00
dtype bf16
device cuda:0
batch size 4, 8
attention implementation flash_attention_2
trust remote code False
max sequence length 32768
candidate ranking reranking_hybrid
rerank top-k not recorded
query prompt name query
document prompt name document
Python 3.12.12 (main, Dec 9 2025, 19:02:36) [Clang 21.1.4 ]
Platform Linux-6.8.0-107-generic-x86_64-with-glibc2.39
torch 2.9.0
transformers 5.12.1
sentence-transformers 5.4.1
datasets 4.8.4
CUDA available=True, version=12.8
CUDA devices 0: NVIDIA GeForce RTX 5090

Command 

CUDA_VISIBLE_DEVICES=1 uv run hakari-bench evaluate from-model-card \
  --model-card config/model_cards/nvidia__Nemotron-3-Embed-1B-BF16.yaml \
  --all \
  --dtype bf16 \
  --attn-implementation flash_attention_2 \
  --batch-size 8 \
  --device cuda:0 \
  --results-dir output/nemotron-3-embed-1b-tf5121-fa283-full

Submitter Notes

  • The model's registered SentenceTransformers prompts were used: query: for queries and passage: for documents. Evaluation used bf16 and the model-recommended flash_attention_2 implementation. No embedding truncation was configured; the native 2,048 dimensions were evaluated.
  • The initial batch size was 8. A long-document task exhausted GPU memory, so the resumed portion (251 tasks) used batch size 4 with PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True; completed task results were reused. Model max sequence length remained the native 32,768 tokens throughout.
  • These are the complete standard --all results: 551 task JSON.XZ files. The leaderboard Overall intentionally deduplicates 13 overlapping task copies and therefore reports 538 raw task results / 369 grouped units.

Checklist

  • Result files are committed under hakari-results/nvidia__Nemotron-3-Embed-1B-BF16/.
  • 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.
hotchpotch changed pull request status to merged

Sign up or log in to comment