MLM-Scaling-datasets

Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
mc_labels: list<item: double>
  child 0, item: double
input: string
n_tokens: int64
-- schema metadata --
huggingface: '{"info": {"features": {"mc_labels": {"feature": {"dtype": "' + 150
to
{'input': Value('string'), 'n_tokens': Value('int64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1872, in _prepare_split_single
                  for key, table in generator:
                                    ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 609, in wrapped
                  for item in generator(*args, **kwargs):
                              ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/arrow/arrow.py", line 74, in _generate_tables
                  yield Key(file_idx, batch_idx), self._cast_table(pa_table)
                                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/arrow/arrow.py", line 54, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2272, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2218, in cast_table_to_schema
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              mc_labels: list<item: double>
                child 0, item: double
              input: string
              n_tokens: int64
              -- schema metadata --
              huggingface: '{"info": {"features": {"mc_labels": {"feature": {"dtype": "' + 150
              to
              {'input': Value('string'), 'n_tokens': Value('int64')}
              because column names don't match
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1342, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 907, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1739, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1922, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

input string	n_tokens int64
CC[C@]C)CCNC=O)[C@H]CCcccccOC))c6))))))=NO5))))))C5	35
Ccccncc6CCC=O)N[C@@H]CC[C@H]5C[C@@H]NC=O)[C@]O)CCNC=O)OCC)C)C))))C5)))))))C7	56
CncC=O)N[C@H]CCNC=O)CCCCCCC5))))))))C5)))))))c[nH]c5=O	38
C[C@@H]COccccF)cc6F)))))))))NC=O)ccc[N+]=O)[O-]))nn5C	38
CCC)C)OC=O)NCcccccNC=O)NCCCCC4)))[C@H]4[C@H]CCCO5)))))))))))c6	40
COcccc[C@H]C[C@H]3C=O)NCC[C@H][C@@H]5CCN5C=O)cccc-n[nH]cC)cc5=O))))))cc6)))))))))))))))))))cc6	70
O=CNCcccccCOCcccco5))))))))c6))))))))[C@@H]C[C@@H]CCCC[C@H]6N9C=O)cccccc6	44
CCOC=O)NCCC[C@H]C=O)NccccNCCOCC6))))))cc6C)))))))))C6	32
O=Cccccnccnc6c%10))))))))))NCCO[C@@]CCNC=O)[C@H]ON=Ccccccc6))))))O5))))))C5))))C6	52
CCOccccCNCC))C=O)CCCNC=O)CC)C)N=C5O))))))))))))cc6OC	31
CCC)CCCC=CC=C5)))))CCC)C)C6	17
CNC)CO)=NN=CO)cccF)cF)cc6Cl	14
O=C[O-])C[C@H]C=O)NCC[NH+]6C[C@H]O)Cccccccccc6c%10	34
CCCcnocC[NH+]C[C@H]C[C@H]5[C@H]4NC=O)cccncc6F))))))))))))))))n5	45
CNCCCO)=NcccC=O)NC[C@@H]CN))[C@H]cccccc6))))))C5))))))ccc6%11	39
Cncc[C@@H]CO))NC=O)C=O)Ncccnn5Cccnn-cccccc6))))))c5))))))))))))))))cn5	45
CC[C@H]C)C=O)N[C@@H]CC[C@H]5[C@@H]NC=O)ccccn[nH]cC)c5c9)))))))))))C6	55
CCC)C)[C@@H]OCCC[C@H]6C[NH2+]Cccnc-cccco5)))))s5	30
CCC)OCC[C@H]O)C[NH+]Ccccc-ncncn5)))))cc6)))))))CCC3	32
CcccC[NH+]C)[C@@H]C[C@@H]CNC=O)OCC)C)C))))C[C@@H]5C8))))))))))nC)n5	54
CNCCSC)=O)=O))))C=O)cccccccc6cc%10OCF)F	20
CCCCNC=O)N[C@@H]cccccNC=O)CCCCCC6))))))))c6))))))CC=O)OC)))=C6C	41
CcccncNC[C@@H]O)[C@H]NC=O)COCC=Ccccccc6)))))))))))))C5)))))n6	41
CNCccnnC)c5))))))C=O)CNCCOC[C@@H]6C#N	23
C=Ccncncc6CCCCC7)))))))))))CCC5	15
CcnocC)c5C[NH+]CC[C@H]CNC=O)CCccnc[nH]5))))))))))C5	33
CCC)CC=O)NCCC[C@@H]CNC=O)CnncnC)c5=O))))))))))C6	33
CCC=O)NCC[C@@H]C)[C@@H]CNC=O)cccCl)cCl)[nH]5))))))))C6	40
CCcnncC[NH2+][C@H]CCO)))cccCl)ccCl)c6)))))))))o5	29
Cncnc-cccccNS=O)=O)cccF)cF)cc6F)))))))))c6))))))n5	31
C[C@@H]CNcnncCNCCCC5=O)))))))n5CCNC=O)CC)C)C)))))))))))C[C@]CCCOC6)))))O6	49
C[NH+]CC[C@@H]C=O)NC[C@@H]CCN4C=O)ccn[nH]c5CCC3))))))))))))))))C5	49
CC[C@H]C)cnncNCCC[C@H]Ocncccc6F))))))))C6))))))n5C[C@@H]CCC=CO6	40
Cnnncc5C[NH+]C[C@H]O)[C@H]NC=O)[C@@H]CCccnnC)c5C9)))))))))))C5	43
C=C[C@@H]OCC[C@H]5C=O)NCC[C@H]CNC=O)ccccnnnnc95))))))))))C[C@H]58	47
CCC)OcccCN=CO)NCCNcccccc6C#N))))))))CC6)))))))))ccn6	28
CcccCCCCO)=NcccC=O)NC)C)))ccc6Cl))))))))))))cC)s5	29
O=C[O-])CCCCNS=O)=O)cccc-cnn[n-]n5)))))cc6Cl))))))))CC6	38
CCcccc=O)cOcccccc6)))))))cC)oc6cC[NH+]C)C)))c%10O	28
COC=O)[C@@H]CCC[C@@H]5S=O)=O)N[C@H]C)COcccccc6CF)F)F	37
NC=O)cccccS[C@@H]CCNcsccc5C=O)[O-])))))))C5=O)))))))c6	37
C=CC)COPCC	5
CCOCCOccCl)cccc6N=CO)CCCO)=Ncccccc6%10	12
CCCS=O)=O)[N-]ccccNC=O)[C@H]CC=CC[C@H]6C=O)[O-]))))))))))cc6	47
CCOcccNC=O)C[C@@H]C=O)NccBr)cccc69))))))))))))ccc6C	31
COC=O)COcccccC=O)Ncnccn5Ccccccc6))))))))))))))c6	24
CCC)COccccS=O)=O)NCCC#N))))CCC)C)C)))))cc6	27
C[C@@H]NC=O)ccccs5)))))))[C@H]CCNC=O)CCC[NH+]C)CC6)))))))C5	44
COccccncC=O)NCCNCCO)))CC6)))))))cC)nc95	19
CcncCOcccccC=O)[C@H]C#N))cnccC)cccc6[nH]9)))))))))))c6))))))))cs5	42
Cccccc-cnncn5C[C@@H]C=O)NCcccccc6)CCO5)))))))))))CC6)))))))))c6	38
O=CNCC[NH+]CCcsccc5C9))))))))))))N[C@H]cccccc6))))))cccccc6)CCCO6	36
C[C@@H]C[C@H]3C=O)NCCNccccSC)=O)=O))cc6[N+]=O)[O-]	38
CccccCC=O)NCcccccNCC[C@H][NH3+])C5)))))n6))))))))))csc5s8	33
CCC)CC[C@H]CCCN5C=O)C=O)NC[C@H]C)cnccs5	25
COCCC)C)C[NH2+]CCcccC)ccc6OC	12
CcccS=O)=O)NCCN=CO)NCCNCCCCC6)))))))))))))))cC)s5	30
CCC)[C@@H][NH2+][C@H]CCCNCC)C))C6=O))))))))C=O)[O-]	41
Ccccnncnc5c9NC=O)NCCOcccc-cnn[n-]n5)))))cc6)))))))C4	26
CCcnc[C@@H]CCCCN6C=O)ccc=O)[nH]ccccF)cc%106)))))))))))))))))no5	41
COcccccC)c6NC=O)C=O)NCCncnnc5CF)F)F)))))C6	21
CccC=O)N[C@@H]CCCN[C@H]C)ccccF)cc6)))))))C6=O)))))))))nnn5CccccF)cc6	44
CNCCC[C@H][NH-])CO)=Ncnccs5))))))))))NS=O)=O)ccC)ccOC=N)))cC)c6C	43
C[C@@H]CNC=O)CC=O)NC)C)))CC3))))C[C@H]5C[NH2+]C[C@H]CCCF)F)C5	48
CC=O)cccNC=O)NCCO[C@@H]C[C@H]37)))))))))ccc6F	32
COC=O)[C@@]C)NC=O)cccSN)=O)=O))cc[N+]=O)[O-]))c6C)))))))))CCC3	49
CccC=O)NCCOC[C@]CNC=O)ccccnnC)cc5c9))))))))))CCO6)))))C7))))))))ncccccn96	42
COcccC)c-cncccC)cSN)=O)=O))cc6n9C))))))))))cn6	27
CCOC=O)CCCCNC=O)N[C@@H]C)ccccBr)cc6))))))))))CC6	31
CCC)CCNC=O)cn[nH]nc5-cccccc6))))))))))))))C3C)C	30
cccC[NH+]CCcn[nH]nc5C9))))))))))ccOCCCCC5))))))c6	28
CC[C@@H]C)CC=O)NCCCC)CNC=O)ccnccF)c6)))))))))CC6	32
CC[C@@H]OCC[C@H]5C=O)NC[C@@H][C@H]C5)[C@H]3NC=O)CC)CCCCCC7	43
CCOCCS=O)=O)NC[C@H]cccccc6Cl)))))))cc[nH]cccccc96)))))))[C@H]6C	39
C[NH+]CCC[C@H]5C=O)NCC[C@H]CO[C@@H]CNC=O)[C@@H]C[C@H]3CCC3)))))))))[C@H]5C9	60
CC[C@@H]CCNC=O)cccccc6SCC=O)NcccccC#N))c6)))))))))))))))))[C@@H]5C	45
COCC=O)NCC[C@@H]C)[C@@H]NC=O)cccccF)ccc6[nH]9)))))))))))C6	42
CNCC[NH+]CCNcccccc6))))))CC6))))))))C=O)NCcccccCN)=O))c6	32
CCC)[NH+]CCC[C@H]Cnc-ccccnc6))))))nnc5NC)CccccBr)cc6))))))))))))))C6	41
O=S=O)Nccccncsc5c9Br)))))))))))cccccc6CF)F)F	23
N#Ccccncc6NCCNS=O)=O)cccc-cccccc6))))))cc6)))))))CC6	27
CCNC=O)[C@@H]C)[NH2+][C@H]C)cccccc6)CCCC6)))))))))))))[C@@H]CCS=O)=O)C5	53
C[C@H][NH2+]CCCO))CCCCC6))))))))cccccc6OCC=O)[O-]	30
CCCcccC=O)NCC[C@H]C)[C@H]C[NH2+]CCF)))))C6)))))))ccn6	37
C=CC[C@H]C)C=O)NC[C@H]CCNC=O)cccnCC)C))n5))))))[C@@H]5ccccnc6	43
CCC)ncccC=O)NC[C@@H]C)[C@H]NC=O)C=CCCCCC7)))))))))C5))))))n5	44
CNC=O)ccccnnccc95))))))))))[C@H]CC[NH+]CC[C@@H]CCCO5)))))))C5	41
COC=O)[C@@H]cccccc6))))))[C@H]CCCC[NH+]6CccncBr)cc6C	30
CC=Ccccccc6))))))[C@H]3CO)=NCCcnocCCC3)))n5	22
COC=O)[C@H]OC))[C@]C)OcccOC))ccc6C=O)[C@H]%10C5	32
CC[C@H]C)NC=O)CCnc-cccccc6OC))))))))nnc5NCCCC=O)NC)))CC6	32
CC=O)O[C@@H]C=O)NccccCNC=O)CncO)csc5=S)))))))))CCC4))))cc6))))))))CC)C	50
CCCOCCNCCCN))OCC)C)C6	8
Cccc-cccc[nH]cC#N))cc5n9)))))))))ccC)c6OCC)C	26
CNC=O)CCNC=O)cccccc6-cncCC)C))no5))))))))))))C4	29
O=CNC[C@H]CCCCS6=O)=O)))))))))NCCO[C@@]CCcccccc69))))))))C6	39
CNCC[NH+]CcccccNC=O)ccccBr)s5)))))))c6)))))))CC6	28
CCC)[C@@H]C=O)NCcccccc6NCC[NH+]C)CC6)))))))))))))))nnnc-cccccc6))))))n5	45
CCCOC=O)COcccccCF)F)F))c6)))))))))))COC4	24
cccc-cccc-cccccc6)-cccccc-cccc-cccccc6))))))cc6))))))ccc-%11c%106)))))))))))))))cc6))))))cc6	54

End of preview.

MLM-Scaling-datasets

Dataset Summary

MLM-Scaling-datasets is the companion dataset repository for the paper "Unveiling Scaling Behaviors in Molecular Language Models: Effects of Model Size, Data, and Representation". It packages the molecular corpora used to study how model size, token budget, and molecular representation affect autoregressive molecular language modeling.

The source molecules are collected from ZINC and UniChem and serialized into multiple molecular string representations. The repository is designed for:

compute-controlled scaling studies
pretraining GPT-style molecular language models
controlled comparison across molecular representations
downstream transfer studies for molecular property prediction

Dataset Source

According to the paper, the pretraining corpus is constructed from large-scale unlabeled molecules collected from:

ZINC
UniChem

Each molecule is then converted into one of the molecular string representations used in the study.

Molecular Representations

The paper studies five string representations:

SMILES
DeepSMILES
SAFE
FragSeq
FragLink

Repository Contents

The current repository layout contains the following subsets.

DeepSMILES

DeepSMILES-100M
DeepSMILES-300M
DeepSMILES-1B
DeepSMILES-3B

FragSeq

FragSeq-100M
FragSeq-300M
FragSeq-1B
FragSeq-3B

FragLink

FragLink-100M
FragLink-300M
FragLink-1B
FragLink-3B

SAFE

SAFE-100M
SAFE-300M
SAFE-1B
SAFE-3B

SMILES

SMILES-100M
SMILES-300M
SMILES-1B
SMILES-3B

What the Scale Labels Mean

The scale labels in the repository names refer to token budget after tokenization under a given representation.

For the main scaling grid in the paper, the following token budgets are used:

100M
300M
1B
3B

This point matters: a "1B-token" subset under one molecular representation is not the same thing as a "1B-token" subset under another representation, because sequence length and token statistics change with the representation.

Dataset Creation and Experimental Role

The paper uses these corpora in a structured scaling study with:

8 model sizes: 1M, 4M, 16M, 43M, 85M, 152M, 278M, 650M parameters
4 token budgets: 100M, 300M, 1B, 3B tokens
5 molecular representations

The main training grid uses single-epoch from-scratch runs for compute-controlled comparison.

Dataset Structure

At a conceptual level, each subset contains molecular strings serialized under a specific representation and scale.

Examples of what varies across subsets:

the molecular string representation
the token budget
the token statistics and sequence lengths
the downstream behavior of models trained on that subset

This repository is therefore best understood as a family of matched corpora rather than a single flat dataset.

Intended Use

This dataset repository is suitable for:

pretraining autoregressive molecular language models
scaling-law studies under matched compute
representation comparison in molecular language modeling
initialization studies for downstream molecular property prediction

Out-of-Scope Use

This repository is not intended to be used as:

a clinical knowledge base
a stand-alone benchmark for safety or efficacy claims
a substitute for chemistry-specific filtering, synthesis planning, docking, or wet-lab validation

How to Download

For a multi-folder release like this one, the most robust way is to download the repository snapshot and then select the subset you need.

from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    repo_id="SZU-ADDG/MLM-Scaling-datasets",
    repo_type="dataset",
)

print(local_dir)

After downloading, choose the folder that matches your target representation and token budget, for example:

SMILES-1B
DeepSMILES-300M
FragSeq-3B
FragLink-1B
SAFE-100M

Companion Resources

Model repository: SZU-ADDG/MLM-Scaling-Model
Code: SZU-ADDG/MLM-Scaling
Paper: arXiv:2601.22757

Citation

If you use this dataset repository in your research, please cite:

@article{xu2026mlmscaling,
  title={Unveiling Scaling Behaviors in Molecular Language Models: Effects of Model Size, Data, and Representation},
  author={Xu, Dong and Pan, Qihua and Yuan, Sisi and Li, Jianqiang and Zhu, Zexuan and Ji, Junkai},
  journal={arXiv preprint arXiv:2601.22757},
  year={2026}
}

Downloads last month: 13

Models trained or fine-tuned on SZU-ADDG/MLM-Scaling-datasets

SZU-ADDG/MLM-Scaling-Model

Updated 1 day ago

Paper for SZU-ADDG/MLM-Scaling-datasets

Unveiling Scaling Behaviors in Molecular Language Models: Effects of Model Size, Data, and Representation

Paper • 2601.22757 • Published Jan 30