name stringlengths 11 50 | hf_repo stringlengths 26 62 ⌀ | stage stringclasses 5
values | kind stringlengths 4 21 | status stringclasses 5
values | maintainer stringclasses 3
values | upstream stringlengths 9 70 ⌀ | decontaminated stringclasses 1
value | tokenizer stringclasses 5
values | rows_internal float64 4.2k 3.14M ⌀ | tokens_internal float64 24.1M 2.36B ⌀ | used_in stringlengths 13 35 ⌀ | notes stringlengths 13 90 ⌀ | visibility stringclasses 2
values | license stringclasses 6
values | n_languages float64 0 70 ⌀ | rows_hub float64 0 8.6M ⌀ | parquet_gb float64 0 189 ⌀ | last_modified stringdate 2025-07-22 00:00:00 2026-11-05 00:00:00 ⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
oellm-longctx-tokenized-natural-128k-256k-pilot-v1 | birgermoell/oellm-longctx-tokenized-natural-128k-256k-pilot-v1 | mid-train | long-context | pilot | long-context (AI Sweden) | null | null | OELLM 256K | null | null | null | null | public | other | 2 | 0 | null | 2026-06-22 |
oellm-longctx-tokenized-streamed-all-v2 | birgermoell/oellm-longctx-tokenized-streamed-all-v2 | mid-train | long-context | released | long-context (AI Sweden) | null | null | OELLM 256K | null | null | oellm-9b-128k | null | public | other | 35 | 0 | null | 2026-05-28 |
oellm-longctx-tokenized-structured-128k-256k-v1 | birgermoell/oellm-longctx-tokenized-structured-128k-256k-v1 | mid-train | long-context | released | long-context (AI Sweden) | null | null | OELLM 256K | null | null | oellm-9b-128k / 256k | null | public | other | 1 | 0 | null | 2026-06-23 |
oellm-longctx-tokenized-superlong-512k-1m-2m-v2 | birgermoell/oellm-longctx-tokenized-superlong-512k-1m-2m-v2 | mid-train | long-context | released | long-context (AI Sweden) | null | null | OELLM 256K | null | null | long-context extension | 512K/1M/2M documents incl. synthetic recall | public | other | 6 | 0 | null | 2026-06-25 |
nemotron-cc-10K-sample-translated | openeurollm/nemotron-cc-10K-sample-translated | mid-train | translated-web | experimental | openeurollm | Nemotron-CC-HQ sample | null | null | null | null | null | 10K translated pretraining documents | public | null | 0 | 450,000 | 0.44 | 2025-07-22 |
terminal-agent-sft-mix | birgermoell/terminal-agent-sft-mix | sft | agentic | released | post-training (AI Sweden) | null | null | null | null | null | null | terminal-agent SFT mixture | public | other | 1 | 47,986 | 0.72 | 2026-11-05 |
deep-search-warmstart-v1 | null | sft | agentic-search | internal | post-training (AI Sweden) | deep-search reference trajectories + stage3-mix-v1 replay | null | OELLM 256K (template v2.1) | 17,083 | 24,093,840 | deep-search warm start (running) | 3,897 graded-correct browsing trajectories + replay | not on the Hub | null | null | null | null | null |
lmsys-chat-1m-decontaminated | openeurollm/lmsys-chat-1m-decontaminated | sft | conversation | released | openeurollm | lmsys/lmsys-chat-1m | yes | null | null | null | null | null | public | null | 0 | 997,682 | 2.6 | 2026-03-29 |
oellm-eu-medical-posttrain-v1 | birgermoell/oellm-eu-medical-posttrain-v1 | sft | domain | released | post-training (AI Sweden) | null | null | null | null | null | null | EU medical post-training | public | other | 0 | 0 | null | 2026-06-26 |
oellm-eu-exam-mcq-v1 | birgermoell/oellm-eu-exam-mcq-v1 | sft | exam | released | post-training (AI Sweden) | null | null | null | null | null | null | European exam multiple-choice | public | other | 0 | 0 | null | 2026-06-23 |
openeurollm-model-identity | openeurollm/openeurollm-model-identity | sft | identity | released | openeurollm | synthetic | null | null | null | null | oellm-9b-256k Stages 1-3 | model identity conversations | public | cc-by-4.0 | 12 | 1,120 | 0 | 2026-09-24 |
Dolci-Instruct-SFT-decontaminated | openeurollm/Dolci-Instruct-SFT-decontaminated | sft | instruction | released | openeurollm | allenai/Dolci-Instruct-SFT | yes | null | null | null | null | null | public | odc-by | 70 | 2,151,287 | 6.82 | 2026-03-22 |
dolci-instruct-sft-tokenized | openeurollm/dolci-instruct-sft-tokenized | sft | instruction | released | openeurollm | allenai/Dolci-Instruct-SFT | null | OLMo-core | null | null | null | pre-tokenized for OLMo-core | public | apache-2.0 | 1 | 0 | null | 2026-02-26 |
Dolci-Instruct-SFT-translated | openeurollm/Dolci-Instruct-SFT-translated | sft | instruction | released | openeurollm | allenai/Dolci-Instruct-SFT | null | null | null | null | oellm-9b-256k Stage 1 | machine-translated | public | apache-2.0 | 12 | 5,937,525 | 8.3 | 2026-06-24 |
EuroMultiSynth-Instruct | openeurollm/EuroMultiSynth-Instruct | sft | instruction | released | openeurollm | synthetic | null | null | null | null | null | single-turn synthetic instruction/response pairs, 35 European languages | public | apache-2.0 | 35 | 8,321,196 | 9.69 | 2026-10-23 |
orca-agentinstruct-1M-v1-decontaminated | openeurollm/orca-agentinstruct-1M-v1-decontaminated | sft | instruction | released | openeurollm | microsoft/orca-agentinstruct-1M-v1 | yes | null | null | null | null | null | public | cdla-permissive-2.0 | 1 | 1,045,047 | 4.63 | 2026-03-29 |
EU-Instruct-Following-Synthetic | openeurollm/EU-Instruct-Following-Synthetic | sft | instruction-following | released | openeurollm | synthetic | null | null | null | null | null | synthetic instruction-following, 11 languages | public | apache-2.0 | 11 | 2,994,552 | 3.14 | 2026-10-23 |
Nemotron-Post-Training-Dataset-v2-decontaminated | openeurollm/Nemotron-Post-Training-Dataset-v2-decontaminated | sft | mixture | released | openeurollm | nvidia/Nemotron-Post-Training-Dataset-v2 | yes | null | null | null | null | null | public | cc-by-4.0 | 6 | 6,339,857 | 97.86 | 2026-03-29 |
open-perfectblend-decontaminated | openeurollm/open-perfectblend-decontaminated | sft | mixture | released | openeurollm | mlabonne/open-perfectblend | yes | null | null | null | null | null | public | apache-2.0 | 0 | 1,418,925 | 2.9 | 2026-03-29 |
OpenEuroLLM-Excellent-SFT | openeurollm/OpenEuroLLM-Excellent-SFT | sft | mixture | released | openeurollm | Dolci, Nemotron, Aya and others | null | null | null | null | oellm-9b-256k Stage 1 (direct rows) | multilingual SFT mixture; its Nemotron multilingual rows are reasoning with English traces | public | other | 1 | 6,187,716 | 27.61 | 2026-09-28 |
smoltalk2-decontaminated | openeurollm/smoltalk2-decontaminated | sft | mixture | released | openeurollm | HuggingFaceTB/smoltalk2 | yes | null | null | null | null | null | public | null | 0 | 8,604,212 | 189.33 | 2026-03-29 |
stage1-mix-v1 | null | sft | mixture | internal | post-training (AI Sweden) | Excellent SFT, Dolci translated, Aya, identity | null | OELLM 256K (template v2) | 3,135,379 | 2,364,573,138 | oellm-9b-256k Stage 1 | pre-tokenized; 1.71B supervised tokens | not on the Hub | null | null | null | null | null |
Dolci-Think-SFT-32B-decontaminated | openeurollm/Dolci-Think-SFT-32B-decontaminated | sft | reasoning | released | openeurollm | allenai/Dolci-Think-SFT-32B | yes | null | null | null | null | null | public | odc-by | 1 | 2,252,837 | 77.47 | 2026-03-22 |
Dolci-Think-SFT-7B-decontaminated | openeurollm/Dolci-Think-SFT-7B-decontaminated | sft | reasoning | released | openeurollm | allenai/Dolci-Think-SFT-7B | yes | null | null | null | null | null | public | null | 0 | 2,267,351 | 77.57 | 2026-03-22 |
dolci-think-sft-tokenized | openeurollm/dolci-think-sft-tokenized | sft | reasoning | released | openeurollm | allenai/Dolci-Think-SFT-7B | null | OLMo-core | null | null | null | pre-tokenized for OLMo-core | public | apache-2.0 | 1 | 0 | null | 2026-02-26 |
Dolci-Think-SFT-translated | openeurollm/Dolci-Think-SFT-translated | sft | reasoning | released | openeurollm | allenai/Dolci-Think-SFT-32B | null | null | null | null | null | translated with gemma-4-31B-it | public | apache-2.0 | 0 | 1,735,919 | 19.96 | 2026-10-28 |
openeurollm-reasoning-instruct-v1 | birgermoell/openeurollm-reasoning-instruct-v1 | sft | reasoning | released | post-training (AI Sweden) | Dolci Think, Nemotron, replay | null | OELLM 256K (template v2) | 1,164,312 | 2,337,478,029 | oellm-9b-256k Stage 2 | reasoning + direct replay | public | other | 36 | 1,166,773 | 2.63 | 2026-10-21 |
reasoning-traces-multilingual | openeurollm/reasoning-traces-multilingual | sft | reasoning | pilot | openeurollm | null | null | null | null | null | null | multilingual math reasoning traces, two-stage pilot | public | cc-by-4.0 | 37 | 3,425 | 0.01 | 2026-08-17 |
oellm-eu-defect-repair-sft-v1 | birgermoell/oellm-eu-defect-repair-sft-v1 | sft | repair | released | post-training (AI Sweden) | null | null | null | null | null | null | defect-repair SFT | public | null | 0 | 57,352 | 0.12 | 2026-06-25 |
function-calling-sft-oellm-64k | openeurollm/function-calling-sft-oellm-64k | sft | tool-use | preliminary | openeurollm | null | null | OELLM 64K | null | null | null | English function-calling SFT, prepared train partition and fixed holdout | public | null | 1 | 8,425,065 | 26.03 | 2026-10-12 |
oellm-eu-tooluse-v1 | openeurollm/oellm-eu-tooluse-v1 | sft | tool-use | released | openeurollm | xLAM, Hermes FC, ToolACE | null | null | null | null | null | Hermes/Qwen tool-call format | public | apache-2.0 | 1 | 55,077 | 0.1 | 2026-09-24 |
openeurollm-tool-use-v1 | birgermoell/openeurollm-tool-use-v1 | sft | tool-use | released | post-training (AI Sweden) | Toucan, Nemotron tool calling, ToolMind, ToolACE, When2Call | null | OELLM 256K (template v2.1) | 158,703 | 327,899,529 | oellm-9b-256k Stage 3 | 16 languages, single/parallel/multi-step calls | public | other | 17 | 158,703 | 0.28 | 2026-10-22 |
stage3-mix-v1 | null | sft | tool-use | internal | post-training (AI Sweden) | openeurollm-tool-use-v1 + Stage-2 replay | null | OELLM 256K (template v2.1) | 285,737 | 410,734,152 | oellm-9b-256k Stage 3 | pre-tokenized | not on the Hub | null | null | null | null | null |
Dolci-Instruct-DPO-translated | openeurollm/Dolci-Instruct-DPO-translated | dpo | preference | released | openeurollm | allenai/Dolci-Instruct-DPO | null | null | null | null | oellm-9b-256k Stage 4 DPO | machine-translated | public | apache-2.0 | 11 | null | null | 2026-09-15 |
dpo-v2-250k | null | dpo | preference | internal | post-training (AI Sweden) | UltraFeedback, HelpSteer3, Dolci-Instruct-DPO-translated, format pairs | null | OELLM 256K (template v2.1) | 250,000 | null | oellm-9b-256k Stage 4 DPO | pre-tokenized 250K-pair subset | not on the Hub | null | null | null | null | null |
oellm-dpo-metadataset | birgermoell/oellm-dpo-metadataset | dpo | preference | released | post-training (AI Sweden) | null | null | null | null | null | null | DPO source catalogue | public | apache-2.0 | 24 | 65 | 0 | 2026-09-14 |
oellm-excellent-dpo-v0.1 | birgermoell/oellm-excellent-dpo-v0.1 | dpo | preference | released | post-training (AI Sweden) | null | null | null | null | null | null | null | public | other | 0 | 73,964 | 0.15 | 2026-10-07 |
safety-dpo-prompts-v1 | null | dpo | safety | internal | post-training (AI Sweden) | null | null | null | 4,200 | null | null | safety DPO prompts by behaviour | not on the Hub | null | null | null | null | null |
deep-search-v1 | null | rl | agentic-search | internal | post-training (AI Sweden) | Wikidata two-hop chains over Wikipedia 20231101 | null | null | 47,826 | null | E10-web (planned) | two-hop browsing QA in 10 languages, snapshot-verified; 7,339 eval rows | not on the Hub | null | null | null | null | null |
oellm-code-rlvr | openeurollm/oellm-code-rlvr | rl | code | released | openeurollm | procedural | null | null | null | null | oellm-9b-256k RL windows | 100K Python prompts with tests | public | apache-2.0 | 1 | 100,000 | 0.09 | 2026-09-24 |
oellm-eu-tasks | birgermoell/oellm-eu-tasks | rl | eu-tasks | released | post-training (AI Sweden) | Wikidata and EU documents | null | null | null | null | oellm-9b-256k RL (E8) | eu_qa / eu_doc suites with +1/0/-1 reward | public | cc-by-4.0 | 10 | 37,891 | 0.01 | 2026-10-27 |
oellm-math-rlvr | openeurollm/oellm-math-rlvr | rl | math | released | openeurollm | procedural | null | null | null | null | oellm-9b-256k RL windows | 1M verifier-ready math problems | public | apache-2.0 | 24 | 1,000,000 | 0.09 | 2026-09-24 |
oellm-rl-pools | birgermoell/oellm-rl-pools | rl | mixture | released | post-training (AI Sweden) | oellm-math-rlvr, oellm-code-rlvr, eu-tasks, tool pools | null | null | null | null | oellm-9b-256k RL windows W1-W3, E8 | RL prompt pools per window | public | other | 10 | 7,168 | 0 | 2026-11-02 |
oellm-eu-eval-holdouts-v1 | birgermoell/oellm-eu-eval-holdouts-v1 | decontamination | holdout | released | post-training (AI Sweden) | null | null | null | null | null | null | held-out eval items for decontamination | public | cc0-1.0 | 38 | 7,600 | 0 | 2026-06-22 |
contaminated-documents | openeurollm/contaminated-documents | decontamination | reference | released | openeurollm | Nemotron, HPLT | null | null | null | null | null | benchmark-contaminated documents found with nemo-curator | public | null | 0 | 40,269 | 0.21 | 2025-11-25 |
OpenEuroLLM post-training data registry
One table of the datasets used for mid-training, SFT, preference tuning (DPO) and RL in OpenEuroLLM, whether they live in this organisation, under a member account, or only on a cluster. Use it to find what already exists before building something new, to see which models were trained on what, and to keep the counts in one place.
The table is registry.csv (also shown in the dataset viewer above). Each row is one dataset.
Columns
| Column | Meaning |
|---|---|
name |
short name |
hf_repo |
Hub repository; empty if the data is not on the Hub yet |
stage |
mid-train, sft, dpo, rl, or decontamination (reference sets used to filter training data) |
kind |
what the data teaches, e.g. instruction, reasoning, tool-use, agentic-search, long-context, preference |
status |
released, preliminary, pilot, experimental, private (on the Hub, not public) or internal (cluster only) |
maintainer |
the group that builds and answers for it |
upstream |
source datasets or generators |
decontaminated |
yes if filtered against evaluation benchmarks; empty if unknown |
tokenizer |
set when the data is pre-tokenized (and the chat template it was rendered with) |
rows_internal, tokens_internal |
counts from the build manifest, for data that is not on the Hub or where the viewer has no count |
used_in |
models or training stages that used it |
notes |
anything else worth knowing |
visibility, license, n_languages, rows_hub, parquet_gb, last_modified |
filled from the Hub by build_registry.py |
Adding or updating an entry
- Edit
entries.csv: add a row, or change the curated columns of an existing one. - Run
HF_TOKEN=... python build_registry.pyto refresh the Hub columns and rewriteregistry.csv. - Open a pull request on this repository with both files.
Please do not put cluster paths, account names or personal data in the table; name the group in maintainer.
- Downloads last month
- 49