antidoom:pubmedqa_artificial_train:23889
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_lmsys_arena:27030
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:16373
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_autoif:36172
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:26863
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_autoif:40402
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:10498
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:pubmedqa_artificial_train:5114
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_ultrainteract:40875
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_ultrainteract:5408
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:42991
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_ultrachat200k:3172
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:1207
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_ultrainteract:3309
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:47489
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:43448
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:28760
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:pubmedqa_artificial_train:15698
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_evol_codealpaca:6933
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:ifstruct_train_generated:19064
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:math_lighteval_train:2860
DigitalLearningGmbH/MATH-lighteval
https://huggingface.co/datasets/DigitalLearningGmbH/MATH-lighteval
counting_and_probability:197
Use train split only; keep test-derived MATH/MATH-500 slices for eval.
antidoom:math_qa_train:8113
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_lmsys_arena:6454
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:16941
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:open_perfectblend_autoif:6808
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:22307
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:pubmedqa_artificial_train:28245
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:ifstruct_train_generated:11673
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:open_perfectblend_ultrachat200k:21233
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:ifstruct_train_generated:5820
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:open_perfectblend_ultrachat200k:21993
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:ifstruct_train_generated:7086
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:mmlu_auxiliary_train:46990
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:open_perfectblend_evol_codealpaca:39689
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:45219
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:ifstruct_train_generated:10187
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:open_perfectblend_ultrachat200k:1395
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:gsm8k_train:3931
https://huggingface.co/datasets/openai/gsm8k
Use train split only; keep GSM8K test for eval.
antidoom:open_perfectblend_metamathqa:49375
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:9338
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_evol_codealpaca:16337
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:37244
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:21883
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_metamathqa:15686
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:33812
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:pubmedqa_artificial_train:21527
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_evol_codealpaca:15695
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:21999
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_autoif:34624
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_evol_codealpaca:30672
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:17596
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_metamathqa:28649
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:pubmedqa_artificial_train:49236
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_ultrainteract:35342
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:37446
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:pubmedqa_artificial_train:22928
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_ultrainteract:5448
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:23489
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:26398
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:pubmedqa_artificial_train:26098
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_evol_codealpaca:25380
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:pubmedqa_artificial_train:40661
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_lmsys_arena:24219
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_metamathqa:22117
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_metamathqa:10296
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:18929
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:open_perfectblend_autoif:14538
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_ultrachat200k:30384
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_ultrachat200k:34963
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:41016
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:pubmedqa_artificial_train:46569
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:ifstruct_train_generated:6251
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:ifstruct_train_generated:4601
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:pubmedqa_artificial_train:48189
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_metamathqa:29979
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:17557
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:open_perfectblend_metamathqa:41452
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:14356
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_metamathqa:26904
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:pubmedqa_artificial_train:33123
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_evol_codealpaca:8588
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_metamathqa:21041
https://huggingface.co/datasets/meta-math/MetaMathQA
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:27429
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:mmlu_auxiliary_train:48394
https://huggingface.co/datasets/cais/mmlu
Use auxiliary_train only; keep MMLU/MMLU-Pro validation and test for eval.
antidoom:pubmedqa_artificial_train:35888
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:math_qa_train:7369
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:math_qa_train:14815
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:open_perfectblend_ultrainteract:10029
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:19326
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_ultrachat200k:41097
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_evol_codealpaca:8825
theblackcat102/evol-codealpaca-v1
https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:25837
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_autoif:8116
Post-training-Data-Flywheel/AutoIF-instruct-61k
https://huggingface.co/datasets/Post-training-Data-Flywheel/AutoIF-instruct-61k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:ifstruct_train_generated:968
LiquidAI/ifstruct_generated
Generated from IFStruct train taxonomies only; reserve test__* taxonomies for eval.
antidoom:open_perfectblend_lmsys_arena:44118
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:math_qa_train:5754
https://huggingface.co/datasets/allenai/math_qa
Use train split only; keep validation/test for eval.
antidoom:pubmedqa_artificial_train:768
https://huggingface.co/datasets/pubmed_qa
Prefer artificial train for volume; avoid public labeled eval rows if reporting PubMedQA.
antidoom:open_perfectblend_ultrainteract:15441
openbmb/UltraInteract_sft
https://huggingface.co/datasets/openbmb/UltraInteract_sft
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_lmsys_arena:29341
mlabonne/lmsys-arena-human-preference-55k-sharegpt
https://huggingface.co/datasets/mlabonne/lmsys-arena-human-preference-55k-sharegpt
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.
antidoom:open_perfectblend_ultrachat200k:18389
mlabonne/ultrachat_200k_sft
https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k
General instruction-tuning mix, not treated as held-out eval. Exclude component sources with known upstream held-out prompt overlap. Preserve component source and license metadata per row.