Upload dataset files
Browse files- data/LICENSE +201 -0
- data/README.md +551 -0
data/LICENSE
ADDED
|
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Apache License
|
| 2 |
+
Version 2.0, January 2004
|
| 3 |
+
http://www.apache.org/licenses/
|
| 4 |
+
|
| 5 |
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
| 6 |
+
|
| 7 |
+
1. Definitions.
|
| 8 |
+
|
| 9 |
+
"License" shall mean the terms and conditions for use, reproduction,
|
| 10 |
+
and distribution as defined by Sections 1 through 9 of this document.
|
| 11 |
+
|
| 12 |
+
"Licensor" shall mean the copyright owner or entity authorized by
|
| 13 |
+
the copyright owner that is granting the License.
|
| 14 |
+
|
| 15 |
+
"Legal Entity" shall mean the union of the acting entity and all
|
| 16 |
+
other entities that control, are controlled by, or are under common
|
| 17 |
+
control with that entity. For the purposes of this definition,
|
| 18 |
+
"control" means (i) the power, direct or indirect, to cause the
|
| 19 |
+
direction or management of such entity, whether by contract or
|
| 20 |
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
| 21 |
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
| 22 |
+
|
| 23 |
+
"You" (or "Your") shall mean an individual or Legal Entity
|
| 24 |
+
exercising permissions granted by this License.
|
| 25 |
+
|
| 26 |
+
"Source" form shall mean the preferred form for making modifications,
|
| 27 |
+
including but not limited to software source code, documentation
|
| 28 |
+
source, and configuration files.
|
| 29 |
+
|
| 30 |
+
"Object" form shall mean any form resulting from mechanical
|
| 31 |
+
transformation or translation of a Source form, including but
|
| 32 |
+
not limited to compiled object code, generated documentation,
|
| 33 |
+
and conversions to other media types.
|
| 34 |
+
|
| 35 |
+
"Work" shall mean the work of authorship, whether in Source or
|
| 36 |
+
Object form, made available under the License, as indicated by a
|
| 37 |
+
copyright notice that is included in or attached to the work
|
| 38 |
+
(an example is provided in the Appendix below).
|
| 39 |
+
|
| 40 |
+
"Derivative Works" shall mean any work, whether in Source or Object
|
| 41 |
+
form, that is based on (or derived from) the Work and for which the
|
| 42 |
+
editorial revisions, annotations, elaborations, or other modifications
|
| 43 |
+
represent, as a whole, an original work of authorship. For the purposes
|
| 44 |
+
of this License, Derivative Works shall not include works that remain
|
| 45 |
+
separable from, or merely link (or bind by name) to the interfaces of,
|
| 46 |
+
the Work and Derivative Works thereof.
|
| 47 |
+
|
| 48 |
+
"Contribution" shall mean any work of authorship, including
|
| 49 |
+
the original version of the Work and any modifications or additions
|
| 50 |
+
to that Work or Derivative Works thereof, that is intentionally
|
| 51 |
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
| 52 |
+
or by an individual or Legal Entity authorized to submit on behalf of
|
| 53 |
+
the copyright owner. For the purposes of this definition, "submitted"
|
| 54 |
+
means any form of electronic, verbal, or written communication sent
|
| 55 |
+
to the Licensor or its representatives, including but not limited to
|
| 56 |
+
communication on electronic mailing lists, source code control systems,
|
| 57 |
+
and issue tracking systems that are managed by, or on behalf of, the
|
| 58 |
+
Licensor for the purpose of discussing and improving the Work, but
|
| 59 |
+
excluding communication that is conspicuously marked or otherwise
|
| 60 |
+
designated in writing by the copyright owner as "Not a Contribution."
|
| 61 |
+
|
| 62 |
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
| 63 |
+
on behalf of whom a Contribution has been received by Licensor and
|
| 64 |
+
subsequently incorporated within the Work.
|
| 65 |
+
|
| 66 |
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
| 67 |
+
this License, each Contributor hereby grants to You a perpetual,
|
| 68 |
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
| 69 |
+
copyright license to reproduce, prepare Derivative Works of,
|
| 70 |
+
publicly display, publicly perform, sublicense, and distribute the
|
| 71 |
+
Work and such Derivative Works in Source or Object form.
|
| 72 |
+
|
| 73 |
+
3. Grant of Patent License. Subject to the terms and conditions of
|
| 74 |
+
this License, each Contributor hereby grants to You a perpetual,
|
| 75 |
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
| 76 |
+
(except as stated in this section) patent license to make, have made,
|
| 77 |
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
| 78 |
+
where such license applies only to those patent claims licensable
|
| 79 |
+
by such Contributor that are necessarily infringed by their
|
| 80 |
+
Contribution(s) alone or by combination of their Contribution(s)
|
| 81 |
+
with the Work to which such Contribution(s) was submitted. If You
|
| 82 |
+
institute patent litigation against any entity (including a
|
| 83 |
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
| 84 |
+
or a Contribution incorporated within the Work constitutes direct
|
| 85 |
+
or contributory patent infringement, then any patent licenses
|
| 86 |
+
granted to You under this License for that Work shall terminate
|
| 87 |
+
as of the date such litigation is filed.
|
| 88 |
+
|
| 89 |
+
4. Redistribution. You may reproduce and distribute copies of the
|
| 90 |
+
Work or Derivative Works thereof in any medium, with or without
|
| 91 |
+
modifications, and in Source or Object form, provided that You
|
| 92 |
+
meet the following conditions:
|
| 93 |
+
|
| 94 |
+
(a) You must give any other recipients of the Work or
|
| 95 |
+
Derivative Works a copy of this License; and
|
| 96 |
+
|
| 97 |
+
(b) You must cause any modified files to carry prominent notices
|
| 98 |
+
stating that You changed the files; and
|
| 99 |
+
|
| 100 |
+
(c) You must retain, in the Source form of any Derivative Works
|
| 101 |
+
that You distribute, all copyright, patent, trademark, and
|
| 102 |
+
attribution notices from the Source form of the Work,
|
| 103 |
+
excluding those notices that do not pertain to any part of
|
| 104 |
+
the Derivative Works; and
|
| 105 |
+
|
| 106 |
+
(d) If the Work includes a "NOTICE" text file as part of its
|
| 107 |
+
distribution, then any Derivative Works that You distribute must
|
| 108 |
+
include a readable copy of the attribution notices contained
|
| 109 |
+
within such NOTICE file, excluding those notices that do not
|
| 110 |
+
pertain to any part of the Derivative Works, in at least one
|
| 111 |
+
of the following places: within a NOTICE text file distributed
|
| 112 |
+
as part of the Derivative Works; within the Source form or
|
| 113 |
+
documentation, if provided along with the Derivative Works; or,
|
| 114 |
+
within a display generated by the Derivative Works, if and
|
| 115 |
+
wherever such third-party notices normally appear. The contents
|
| 116 |
+
of the NOTICE file are for informational purposes only and
|
| 117 |
+
do not modify the License. You may add Your own attribution
|
| 118 |
+
notices within Derivative Works that You distribute, alongside
|
| 119 |
+
or as an addendum to the NOTICE text from the Work, provided
|
| 120 |
+
that such additional attribution notices cannot be construed
|
| 121 |
+
as modifying the License.
|
| 122 |
+
|
| 123 |
+
You may add Your own copyright statement to Your modifications and
|
| 124 |
+
may provide additional or different license terms and conditions
|
| 125 |
+
for use, reproduction, or distribution of Your modifications, or
|
| 126 |
+
for any such Derivative Works as a whole, provided Your use,
|
| 127 |
+
reproduction, and distribution of the Work otherwise complies with
|
| 128 |
+
the conditions stated in this License.
|
| 129 |
+
|
| 130 |
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
| 131 |
+
any Contribution intentionally submitted for inclusion in the Work
|
| 132 |
+
by You to the Licensor shall be under the terms and conditions of
|
| 133 |
+
this License, without any additional terms or conditions.
|
| 134 |
+
Notwithstanding the above, nothing herein shall supersede or modify
|
| 135 |
+
the terms of any separate license agreement you may have executed
|
| 136 |
+
with Licensor regarding such Contributions.
|
| 137 |
+
|
| 138 |
+
6. Trademarks. This License does not grant permission to use the trade
|
| 139 |
+
names, trademarks, service marks, or product names of the Licensor,
|
| 140 |
+
except as required for reasonable and customary use in describing the
|
| 141 |
+
origin of the Work and reproducing the content of the NOTICE file.
|
| 142 |
+
|
| 143 |
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
| 144 |
+
agreed to in writing, Licensor provides the Work (and each
|
| 145 |
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
| 146 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
| 147 |
+
implied, including, without limitation, any warranties or conditions
|
| 148 |
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
| 149 |
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
| 150 |
+
appropriateness of using or redistributing the Work and assume any
|
| 151 |
+
risks associated with Your exercise of permissions under this License.
|
| 152 |
+
|
| 153 |
+
8. Limitation of Liability. In no event and under no legal theory,
|
| 154 |
+
whether in tort (including negligence), contract, or otherwise,
|
| 155 |
+
unless required by applicable law (such as deliberate and grossly
|
| 156 |
+
negligent acts) or agreed to in writing, shall any Contributor be
|
| 157 |
+
liable to You for damages, including any direct, indirect, special,
|
| 158 |
+
incidental, or consequential damages of any character arising as a
|
| 159 |
+
result of this License or out of the use or inability to use the
|
| 160 |
+
Work (including but not limited to damages for loss of goodwill,
|
| 161 |
+
work stoppage, computer failure or malfunction, or any and all
|
| 162 |
+
other commercial damages or losses), even if such Contributor
|
| 163 |
+
has been advised of the possibility of such damages.
|
| 164 |
+
|
| 165 |
+
9. Accepting Warranty or Additional Liability. While redistributing
|
| 166 |
+
the Work or Derivative Works thereof, You may choose to offer,
|
| 167 |
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
| 168 |
+
or other liability obligations and/or rights consistent with this
|
| 169 |
+
License. However, in accepting such obligations, You may act only
|
| 170 |
+
on Your own behalf and on Your sole responsibility, not on behalf
|
| 171 |
+
of any other Contributor, and only if You agree to indemnify,
|
| 172 |
+
defend, and hold each Contributor harmless for any liability
|
| 173 |
+
incurred by, or claims asserted against, such Contributor by reason
|
| 174 |
+
of your accepting any such warranty or additional liability.
|
| 175 |
+
|
| 176 |
+
END OF TERMS AND CONDITIONS
|
| 177 |
+
|
| 178 |
+
APPENDIX: How to apply the Apache License to your work.
|
| 179 |
+
|
| 180 |
+
To apply the Apache License to your work, attach the following
|
| 181 |
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
| 182 |
+
replaced with your own identifying information. (Don't include
|
| 183 |
+
the brackets!) The text should be enclosed in the appropriate
|
| 184 |
+
comment syntax for the file format. We also recommend that a
|
| 185 |
+
file or class name and description of purpose be included on the
|
| 186 |
+
same "printed page" as the copyright notice for easier
|
| 187 |
+
identification within third-party archives.
|
| 188 |
+
|
| 189 |
+
Copyright 2026 gyunggyung
|
| 190 |
+
|
| 191 |
+
Licensed under the Apache License, Version 2.0 (the "License");
|
| 192 |
+
you may not use this file except in compliance with the License.
|
| 193 |
+
You may obtain a copy of the License at
|
| 194 |
+
|
| 195 |
+
http://www.apache.org/licenses/LICENSE-2.0
|
| 196 |
+
|
| 197 |
+
Unless required by applicable law or agreed to in writing, software
|
| 198 |
+
distributed under the License is distributed on an "AS IS" BASIS,
|
| 199 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 200 |
+
See the License for the specific language governing permissions and
|
| 201 |
+
limitations under the License.
|
data/README.md
ADDED
|
@@ -0,0 +1,551 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# 🗃️ LLM-KO-Datasets
|
| 2 |
+
|
| 3 |
+
> **목표**: Pre-training, Mid-training (Continued Pre-training), Post-training (SFT/RLHF/DPO)에 필요한 **한국어 + 영어 + 다국어 고품질 데이터셋**을 구축합니다.
|
| 4 |
+
>
|
| 5 |
+
> 허깅페이스에서 바로 사용할 수 있는 데이터셋을 **1순위**로 선정했습니다.
|
| 6 |
+
>
|
| 7 |
+
> 💡 **무료로 데이터셋 구축**: 구글 번역기 등 무료 번역 도구를 활용하여 영어 데이터를 한국어로 번역하는 전략도 포함합니다.
|
| 8 |
+
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
## 📚 목차
|
| 12 |
+
- [Pre-training 데이터셋](#pre-training-데이터셋)
|
| 13 |
+
- [영어 (English)](#영어-english)
|
| 14 |
+
- [한국어 (Korean)](#한국어-korean)
|
| 15 |
+
- [Mid-training / Continued Pre-training](#-mid-training--continued-pre-training)
|
| 16 |
+
- [다국어 / CoT 데이터셋](#다국어--cot-데이터셋)
|
| 17 |
+
- [Post-training 데이터셋](#post-training-데이터셋)
|
| 18 |
+
- [SFT (Supervised Fine-Tuning)](#sft-supervised-fine-tuning)
|
| 19 |
+
- [DPO / Preference 데이터셋](#dpo--preference-데이터셋)
|
| 20 |
+
- [RLHF / RM 데이터셋](#rlhf--rm-데이터셋)
|
| 21 |
+
- [무료 번역 전략](#무료-번역-전략-영어---한국어)
|
| 22 |
+
- [평가용 데이터셋](#평가용-데이터셋)
|
| 23 |
+
- [유용한 컬렉션](#유용한-컬렉션)
|
| 24 |
+
- [참고 논문](#참고-논문)
|
| 25 |
+
- [참고 자료](#참고-자료)
|
| 26 |
+
|
| 27 |
+
---
|
| 28 |
+
|
| 29 |
+
## Pre-training 데이터셋
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
### 영어 (English)
|
| 33 |
+
|
| 34 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 35 |
+
|------|------|------|----------|------|
|
| 36 |
+
| **FineWeb** | 15T tokens (45TB) | HuggingFace에서 96개 CommonCrawl 스냅샷을 정제한 **최고 품질** 영어 웹 데이터. 2024년 릴리즈. | ODC-BY 1.0 | [🤗 HuggingFaceFW/fineweb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) |
|
| 37 |
+
| **FineWeb-Edu** | 1.3T tokens | FineWeb에서 **교육적 콘텐츠**만 필터링한 서브셋. SmolLM 학습에 사용됨. | ODC-BY 1.0 | [🤗 HuggingFaceFW/fineweb-edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) |
|
| 38 |
+
| **RedPajama-V2** | 30T tokens | Together AI의 5개 언어 웹 데이터. 84개 CommonCrawl + 40개 품질 어노테이션 제공. | Apache 2.0 | [🤗 togethercomputer/RedPajama-Data-V2](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-V2) |
|
| 39 |
+
| **DCLM-Baseline** | 4T tokens | DataComp-LM의 고품질 필터링 데이터셋. 240T 원본에서 정제됨. | MIT | [🤗 mlfoundations/dclm-baseline-1.0](https://huggingface.co/datasets/mlfoundations/dclm-baseline-1.0) |
|
| 40 |
+
| **Dolma** | 3T tokens | AI2 OLMo 학습용 데이터. 웹, 학술논문, 코드, 책 포함. | ODC-BY | [🤗 allenai/dolma](https://huggingface.co/datasets/allenai/dolma) |
|
| 41 |
+
| **SmolLM-Corpus** | 600B tokens | SmolLM 학습용 경량 코퍼스. Cosmopedia v2 + FineWeb-Edu + Python-Edu 혼합. | Apache 2.0 | [🤗 HuggingFaceTB/smollm-corpus](https://huggingface.co/datasets/HuggingFaceTB/smollm-corpus) |
|
| 42 |
+
| **The Stack v2** | 3B+ files | 600개 언어 코드 데이터. 코드 LLM 학습 필수. | 다양함 | [🤗 bigcode/the-stack-v2](https://huggingface.co/datasets/bigcode/the-stack-v2) |
|
| 43 |
+
|
| 44 |
+
#### 수학/과학 Pre-training 데이터셋 (VAETKI 모델 사용) ⭐
|
| 45 |
+
> 📦 **NC-AI VAETKI 100B 모델** Pre-training에 사용된 고품질 수학/과학 데이터셋입니다.
|
| 46 |
+
|
| 47 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 48 |
+
|------|------|------|----------|------|
|
| 49 |
+
| **FineWeb-2** | 3T+ words | 96개 CommonCrawl 스냅샷 기반 **1000개 이상 언어** 지원. FineWeb의 다국어 버전. VAETKI 한국어 54.5B 토큰 사용. | ODC-BY 1.0 | [🤗 HuggingFaceFW/fineweb-2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2) |
|
| 50 |
+
| **FineWeb2-HQ** | Top 10% | FineWeb2의 **고품질 필터링 서브셋**. XLM-RoBERTa 분류기로 상위 10% 문서만 선택. 6배 빠른 학습 효과. | ODC-BY 1.0 | [🤗 epfml/FineWeb2-HQ](https://huggingface.co/datasets/epfml/FineWeb2-HQ) |
|
| 51 |
+
| **FineMath** | 34B~54B tokens | CommonCrawl에서 필터링한 **수학 교육 콘텐츠**. Markdown/LaTeX 형식. GSM8k/MATH 성능 향상. | ODC-BY 1.0 | [🤗 HuggingFaceTB/finemath](https://huggingface.co/datasets/HuggingFaceTB/finemath) |
|
| 52 |
+
| **proof-pile-2** | 28B+ tokens | Llemma 학습용 **수학 증명 데이터**. ArXiv + AlgebraicStack + OpenWebMath 포함. | 다양함 | [🤗 EleutherAI/proof-pile-2](https://huggingface.co/datasets/EleutherAI/proof-pile-2) |
|
| 53 |
+
| **MegaMath** | 300B+ tokens | LLM360 프로젝트의 **대규모 수학 코퍼스**. 웹/코드/합성 데이터 통합. | Apache 2.0 | [🤗 LLM360/MegaMath](https://huggingface.co/datasets/LLM360/MegaMath) |
|
| 54 |
+
| **Stack-Edu** | 125B tokens | The Stack v2에서 **교육적 코드**만 필터링. FineWeb-Edu와 동일 방법론. MultiPL-E 성능 향상. | Apache 2.0 | [🤗 HuggingFaceTB/stack-edu](https://huggingface.co/datasets/HuggingFaceTB/stack-edu) |
|
| 55 |
+
| **StackExchange_Mar2023** | 52.7GB | StackExchange 전체 Q&A 데이터 (2023년 3월). 기술 지식 풍부. | CC BY-SA | [🤗 HuggingFaceGECLM/StackExchange_Mar2023](https://huggingface.co/datasets/HuggingFaceGECLM/StackExchange_Mar2023) |
|
| 56 |
+
|
| 57 |
+
#### 🚀 NVIDIA Nemotron Pre-training Datasets (2025 최신) ⭐
|
| 58 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 59 |
+
|------|------|------|----------|------|
|
| 60 |
+
| **Nemotron-CC-v2.1** | 3.8B docs | Nemotron 모델 학습용 **최고 품질** CommonCrawl 정제 데이터. | NVIDIA License | [🤗 nvidia/Nemotron-CC-v2.1](https://huggingface.co/datasets/nvidia/Nemotron-CC-v2.1) |
|
| 61 |
+
| **Nemotron-CC-v2** | 8.79B docs | Nemotron CC 대용량 버전. | NVIDIA License | [🤗 nvidia/Nemotron-CC-v2](https://huggingface.co/datasets/nvidia/Nemotron-CC-v2) |
|
| 62 |
+
| **Nemotron-CC-Math-v1** | 190M docs | **133B 토큰** 규모 고품질 수학 Pre-training 데이터. | NVIDIA License | [🤗 nvidia/Nemotron-CC-Math-v1](https://huggingface.co/datasets/nvidia/Nemotron-CC-Math-v1) |
|
| 63 |
+
| **Nemotron-CC-Code-v1** | 216M docs | CommonCrawl 기반 코드 데이터. | NVIDIA License | [🤗 nvidia/Nemotron-CC-Code-v1](https://huggingface.co/datasets/nvidia/Nemotron-CC-Code-v1) |
|
| 64 |
+
| **Nemotron-Pretraining-Code-v2** | 836M docs | 코드 Pre-training 데이터 v2. | NVIDIA License | [🤗 nvidia/Nemotron-Pretraining-Code-v2](https://huggingface.co/datasets/nvidia/Nemotron-Pretraining-Code-v2) |
|
| 65 |
+
| **Nemotron-Pretraining-Specialized-v1** | 60.7M docs | 전문 도메인 Pre-training 데이터. | NVIDIA License | [🤗 nvidia/Nemotron-Pretraining-Specialized-v1](https://huggingface.co/datasets/nvidia/Nemotron-Pretraining-Specialized-v1) |
|
| 66 |
+
| **Nemotron-Pretraining-SFT-v1** | 299M docs | Pre-training 단계 SFT 데이터. | NVIDIA License | [🤗 nvidia/Nemotron-Pretraining-SFT-v1](https://huggingface.co/datasets/nvidia/Nemotron-Pretraining-SFT-v1) |
|
| 67 |
+
| **Nemotron-PrismMath** | 1M pairs | Prismatic Synthesis로 생성한 **다양한 수학 문제-풀이 쌍**. RL 학습용 기반 데이터. | CC BY 4.0 | [🤗 nvidia/Nemotron-PrismMath](https://huggingface.co/datasets/nvidia/Nemotron-PrismMath) |
|
| 68 |
+
| **OpenScience** | 6M pairs | STEM/법/경제/인문 등 **다분야 합성 QA 데이터**. GPQA-Diamond, MMLU-Pro 성능 향상용. | CC BY 4.0 | [🤗 nvidia/OpenScience](https://huggingface.co/datasets/nvidia/OpenScience) |
|
| 69 |
+
| **OpenCodeGeneticInstruct** | 15M+ | Genetic-Instruct 방식으로 생성한 **Python 코딩 instruction**. 코드 생성 능력 향상. | CC BY 4.0 | [🤗 nvidia/OpenCodeGeneticInstruct](https://huggingface.co/datasets/nvidia/OpenCodeGeneticInstruct) |
|
| 70 |
+
|
| 71 |
+
> 📦 **NVIDIA Nemotron Collection**: [🤗 nvidia/Nemotron-Pre-Training-Datasets](https://huggingface.co/collections/nvidia/nemotron-pre-training-datasets)
|
| 72 |
+
|
| 73 |
+
#### 🔬 Allen AI OLMo 3 Pre-training Datasets (2025 최신) ⭐
|
| 74 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 75 |
+
|------|------|------|----------|------|
|
| 76 |
+
| **Dolma3 Pool** | 56.2M docs | OLMo 3 7B Pre-training 전체 데이터 풀. | ODC-BY | [🤗 allenai/dolma3_pool](https://huggingface.co/datasets/allenai/dolma3_pool) |
|
| 77 |
+
| **Dolma3 Mix 6T** | 6T tokens | OLMo 3 7B 학습에 사용된 **전체 데이터 믹스**. | ODC-BY | [🤗 allenai/dolma3_mix-6T-1025-7B](https://huggingface.co/datasets/allenai/dolma3_mix-6T-1025-7B) |
|
| 78 |
+
| **Dolma3 Mix 150B** | 150B tokens | OLMo 3 Pre-training 서브셋. | ODC-BY | [🤗 allenai/dolma3_mix-150B-1025](https://huggingface.co/datasets/allenai/dolma3_mix-150B-1025) |
|
| 79 |
+
|
| 80 |
+
> 📦 **OLMo 3 Pre-training Collection**: [🤗 allenai/Olmo-3-Pre-training](https://huggingface.co/collections/allenai/olmo-3-pre-training)
|
| 81 |
+
|
| 82 |
+
### 한국어 (Korean)
|
| 83 |
+
|
| 84 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 85 |
+
|------|------|------|----------|------|
|
| 86 |
+
| **Korean Wikipedia (2024)** | ~500MB | 2024년 5월 덤프 기준 한국어 위키피디아 전문. Pre-training 기본 데이터. | CC BY-SA | [🤗 lcw99/wikipedia-korean-20240501](https://huggingface.co/datasets/lcw99/wikipedia-korean-20240501) |
|
| 87 |
+
| **Korean Wikipedia Edu** | 필터링 | 교육적 내용 필터링된 한국어 위키피디아. | CC BY-SA | [🤗 devngho/korean-wikipedia-edu](https://huggingface.co/datasets/devngho/korean-wikipedia-edu) |
|
| 88 |
+
| **kowikitext** | ~100MB | 한국어 위키피디아 텍스트 정제 버전. | CC BY-SA | [🤗 heegyu/kowikitext](https://huggingface.co/datasets/heegyu/kowikitext) |
|
| 89 |
+
| **Namuwiki Dataset** | 대용량 | 나무위키 덤프 데이터 (Alpaca 형식이지만 지식 추출용으로 Pre-training 활용 가능). | 비상업적 | [🤗 psymon/namuwiki_alpaca_dataset](https://huggingface.co/datasets/psymon/namuwiki_alpaca_dataset) |
|
| 90 |
+
| **WanJuan-Korean** | 280GB+ | OpenDataLab의 **대규모 한국어 코퍼스**. 7개 대분류, 34개 소분류. 역사/정치/문화/백과 등 포함. VAETKI 68.9B 토큰 사용. | CC BY 4.0 | [🤗 opendatalab/WanJuan-Korean](https://huggingface.co/datasets/opendatalab/WanJuan-Korean) |
|
| 91 |
+
|
| 92 |
+
#### 📝 한국어 합성/교과서 데이터셋 (허깅페이스에서 바로 사용 가능) ⭐
|
| 93 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 94 |
+
|------|------|------|----------|------|
|
| 95 |
+
| **korean_textbooks** | 1~10M | Gemini Pro로 생성한 **한국어 합성 교과서**. "Textbooks Are All You Need" 방법론. | - | [🤗 maywell/korean_textbooks](https://huggingface.co/datasets/maywell/korean_textbooks) |
|
| 96 |
+
| **korean-textbooks-edu** | - | 교육적 한국어 교과서 데이터. | - | [🤗 devngho/korean-textbooks-edu](https://huggingface.co/datasets/devngho/korean-textbooks-edu) |
|
| 97 |
+
| **KOREAN-SyntheticText-1.5B** | 1.5B | HAERAE-HUB 한국어 합성 텍스트. Pre-training용. | - | [🤗 HAERAE-HUB/KOREAN-SyntheticText-1.5B](https://huggingface.co/datasets/HAERAE-HUB/KOREAN-SyntheticText-1.5B) |
|
| 98 |
+
| **ko_llm_annotations v3** | - | 한국어 LLM 합성 데이터. 2024년 9월 업데이트. | - | [🤗 devngho/ko_llm_annotations](https://huggingface.co/datasets/devngho/ko_llm_annotations) |
|
| 99 |
+
| **korean-webtext-edu** | 128만 docs | KOREAN-WEBTEXT에서 **교육적 콘텐츠 필터링**. Qwen3-next-80b-a3b로 점수 산정. | MIT | [🤗 eliceai/korean-webtext-edu](https://huggingface.co/datasets/eliceai/korean-webtext-edu) |
|
| 100 |
+
| **korean-fineweb-edu-demo** | 5% 샘플 | FineWeb-Edu **한국어 번역** 데모 버전. LLM 학습용 교육 텍스트. | MIT | [🤗 eliceai/korean-fineweb-edu-demo](https://huggingface.co/datasets/eliceai/korean-fineweb-edu-demo) |
|
| 101 |
+
|
| 102 |
+
#### 🚀 KORMo-Team 대규모 한국어 데이터셋 (2025 최신) ⭐⭐
|
| 103 |
+
> 📦 **KORMo (Korean Open Reasoning Model)** 프로젝트에서 공개한 대규모 한국어 데이터셋입니다.
|
| 104 |
+
> [📜 논문: arXiv:2510.09426](https://arxiv.org/abs/2510.09426)
|
| 105 |
+
|
| 106 |
+
| 이름 | 크기 | 설명 | 용도 | 링크 |
|
| 107 |
+
|------|------|------|------|------|
|
| 108 |
+
| **korean-web-collection** | 대용량 | 한국어 웹 수집 데이터. KORMo-10B Pre-training용. | Pre-training | [🤗 KORMo-Team/korean-web-collection](https://huggingface.co/datasets/KORMo-Team/korean-web-collection) |
|
| 109 |
+
| **korean-public-corpus** | 대용량 | 한국어 공공 코퍼스. | Pre-training | [🤗 KORMo-Team/korean-public-corpus](https://huggingface.co/datasets/KORMo-Team/korean-public-corpus) |
|
| 110 |
+
| **Kor-CC-Resili-Parsed** | 대용량 | 한국어 Common Crawl 정제 데이터. | Pre-training | [🤗 KORMo-Team/Kor-CC-Resili-Parsed](https://huggingface.co/datasets/KORMo-Team/Kor-CC-Resili-Parsed) |
|
| 111 |
+
| **UltraFineWeb-ko-synth** | 1.13k likes | 한국어 UltraFineWeb 합성 데이터. | Pre-training | [🤗 KORMo-Team/UltraFineWeb-ko-synth](https://huggingface.co/datasets/KORMo-Team/UltraFineWeb-ko-synth) |
|
| 112 |
+
| **FineWeb2-ko-synth** | 644 likes | FineWeb2 한국어 합성 버전. | Pre-training | [🤗 KORMo-Team/FineWeb2-ko-synth](https://huggingface.co/datasets/KORMo-Team/FineWeb2-ko-synth) |
|
| 113 |
+
| **Cosmopedia-ko-synth** | 949 likes | Cosmopedia 한국어 합성 버전. 교과서 스타일. | Mid-training | [🤗 KORMo-Team/Cosmopedia-ko-synth](https://huggingface.co/datasets/KORMo-Team/Cosmopedia-ko-synth) |
|
| 114 |
+
| **NemoPost-ko-synth** | 386 likes | Nemotron Post-training 스타일 한국어 합성. | Mid-training | [🤗 KORMo-Team/NemoPost-ko-synth](https://huggingface.co/datasets/KORMo-Team/NemoPost-ko-synth) |
|
| 115 |
+
| **NemoPost-ko-translated** | 285 likes | Nemotron 데이터 한국어 번역. | Mid-training | [🤗 KORMo-Team/NemoPost-ko-translated](https://huggingface.co/datasets/KORMo-Team/NemoPost-ko-translated) |
|
| 116 |
+
| **IF-bilingual-sft** | 141 likes | 한영 이중언어 SFT 데이터. | SFT | [🤗 KORMo-Team/IF-bilingual-sft](https://huggingface.co/datasets/KORMo-Team/IF-bilingual-sft) |
|
| 117 |
+
| **NemoPost-ko-synth-sft** | 225 likes | SFT용 Nemotron 스타일 데이터. | SFT | [🤗 KORMo-Team/NemoPost-ko-synth-sft](https://huggingface.co/datasets/KORMo-Team/NemoPost-ko-synth-sft) |
|
| 118 |
+
| **preference-dataset-qwen3** | 115 likes | Qwen3 기반 DPO/Preference 데이터. | DPO | [🤗 KORMo-Team/preference-dataset-qwen3](https://huggingface.co/datasets/KORMo-Team/preference-dataset-qwen3) |
|
| 119 |
+
|
| 120 |
+
> 📦 **KORMo 컬렉션**:
|
| 121 |
+
> - [Pre-training Datasets](https://huggingface.co/collections/KORMo-Team/kormo-pretraining-datasets) (14개)
|
| 122 |
+
> - [Mid-training Datasets](https://huggingface.co/collections/KORMo-Team/kormo-midtraining-datasets) (7개)
|
| 123 |
+
> - [SFT Datasets](https://huggingface.co/collections/KORMo-Team/kormo-sft-datasets) (5개)
|
| 124 |
+
|
| 125 |
+
#### 🌐 한영 번역/병렬 말뭉치 (Pre-training 활용 가능) ⭐
|
| 126 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 127 |
+
|------|------|------|----------|------|
|
| 128 |
+
| **aihub-en-ko-translation-12m** | 12M | 10개 AI Hub 번역 데이터셋 통합. 일상/기술/방송/특허 등. | - | [🤗 nayohan/aihub-en-ko-translation-12m](https://huggingface.co/datasets/nayohan/aihub-en-ko-translation-12m) |
|
| 129 |
+
|
| 130 |
+
#### 한국어 코드 데이터셋
|
| 131 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 132 |
+
|------|------|------|----------|------|
|
| 133 |
+
| **korea-university-programming-dataset** | - | 한국 대학 프로그래밍 데이터셋. | - | [🤗 team-monolith/korea-university-programming-dataset](https://huggingface.co/datasets/team-monolith/korea-university-programming-dataset) |
|
| 134 |
+
|
| 135 |
+
> 💡 **TIP**: 한국어 Pre-training 데이터가 부족할 경우, **Post-training 데이터(SFT)의 일부를 Pre-training에 활용**해도 괜찮습니다.
|
| 136 |
+
> - KoCommercial-Dataset (1.44M), koVast (685K) 등은 대화 형식이지만 한국어 지식이 풍부합니다.
|
| 137 |
+
> - Pre-training 단계에서 일부 포함��고, SFT에서 중복 사용해도 무방합니다.
|
| 138 |
+
|
| 139 |
+
---
|
| 140 |
+
|
| 141 |
+
## Mid-training / Continued Pre-training
|
| 142 |
+
|
| 143 |
+
> Mid-training은 Pre-training 이후, SFT 이전에 **도메인 적응** 또는 **언어 적응**을 위해 수행합니다.
|
| 144 |
+
> 한국어 LLM 개발 시 영어 모델을 한국어에 적응시키는 데 주로 사용됩니다.
|
| 145 |
+
|
| 146 |
+
| 이름 | 크기 | 설명 | 용도 | 라이센스 | 링크 |
|
| 147 |
+
|------|------|------|------|----------|------|
|
| 148 |
+
| **Korean Wikipedia + Namuwiki Mix** | - | 위키피디아 + 나무위키 혼합. 한국어 지식 주입용. | 언어 적응 | CC BY-SA | 위 데이터 조합 |
|
| 149 |
+
| **YuLan-Mini Before Annealing** | 2.4B params | 중간 체크포인트. LR annealing 실험용. | Annealing 실험 | Apache 2.0 | [🤗 yulan-team/YuLan-Mini-Before-Annealing](https://huggingface.co/yulan-team/YuLan-Mini-Before-Annealing) |
|
| 150 |
+
| **Korean Textbooks** | - | 한국어 교과서 데이터. 교육적 텍스트. | 도메인 적응 | 확인 필요 | [🤗 Search "korean textbooks"](https://huggingface.co/datasets?search=korean+textbooks) |
|
| 151 |
+
|
| 152 |
+
#### 🔬 OLMo 3 Mid-training (Dolmino) Datasets ⭐
|
| 153 |
+
| 이름 | 크기 | 설명 | 용도 | 라이센스 | 링크 |
|
| 154 |
+
|------|------|------|------|----------|------|
|
| 155 |
+
| **Dolma3 Dolmino Pool** | - | OLMo 3 7B **Mid-training용 전체 데이터 풀**. | Mid-training | ODC-BY | [🤗 allenai/dolma3_dolmino_pool](https://huggingface.co/datasets/allenai/dolma3_dolmino_pool) |
|
| 156 |
+
| **Dolma3 Dolmino Mix 100B** | 100B tokens | OLMo 3 7B Mid-training 믹스 데이터. | Mid-training | ODC-BY | [🤗 allenai/dolma3_dolmino_mix-100B-1025](https://huggingface.co/datasets/allenai/dolma3_dolmino_mix-100B-1025) |
|
| 157 |
+
| **Dolma3 Dolmino Mix 10B** | 10B tokens | Mid-training 소규모 버전. 실험용. | Mid-training | ODC-BY | [🤗 allenai/dolma3_dolmino_mix-10B-1025](https://huggingface.co/datasets/allenai/dolma3_dolmino_mix-10B-1025) |
|
| 158 |
+
| **Dolma3 Longmino Pool** | - | OLMo 3 7B **Long Context** 학습용 풀. | Long Context | ODC-BY | [🤗 allenai/dolma3_longmino_pool](https://huggingface.co/datasets/allenai/dolma3_longmino_pool) |
|
| 159 |
+
| **Dolma3 Longmino Mix 50B** | 50B tokens | Long Context Mid-training 믹스. | Long Context | ODC-BY | [🤗 allenai/dolma3_longmino_mix-50B-1025](https://huggingface.co/datasets/allenai/dolma3_longmino_mix-50B-1025) |
|
| 160 |
+
|
| 161 |
+
> 📦 **OLMo 3 Pre-training Collection**: [🤗 allenai/Olmo-3-Pre-training](https://huggingface.co/collections/allenai/olmo-3-pre-training)
|
| 162 |
+
|
| 163 |
+
---
|
| 164 |
+
|
| 165 |
+
## 다국어 / CoT 데이터셋
|
| 166 |
+
|
| 167 |
+
> **Chain-of-Thought (CoT)** 데이터는 LLM의 추론 능력을 향상시키는 핵심 요소입니다.
|
| 168 |
+
>
|
| 169 |
+
> 다국어 CoT 데이터를 활용하면 한국어 추론 능력도 함께 향상됩니다.
|
| 170 |
+
|
| 171 |
+
### 한국어 추론 데이터셋 ⭐
|
| 172 |
+
|
| 173 |
+
| 이름 | 크기 | 설명 | 링크 |
|
| 174 |
+
|------|------|------|------|
|
| 175 |
+
| **Yi-Sang (KOREAson)** | 5.79M prompts + 3.7M traces | 한국어 네이티브 추론 데이터셋. 웹 Q&A, 시험, STEM, 코드 포함. **가장 큰 한국어 추론 데이터**. | [🤗 KOREAson Collection](https://huggingface.co/collections/KOREAson) |
|
| 176 |
+
| **ko-limo** | 1K | LIMO 논문 데이터 한국어 번역. 추론 능력 강화용. | [🤗 junnei/ko-limo](https://huggingface.co/datasets/junnei/ko-limo) |
|
| 177 |
+
| **NuminaMath-CoT-Ko** | 860K | NuminaMath 한국어 번역. 수학 추론. CC BY-NC 4.0 | [🤗 ChuGyouk/AI-MO-NuminaMath-CoT-Ko](https://huggingface.co/datasets/ChuGyouk/AI-MO-NuminaMath-CoT-Ko) |
|
| 178 |
+
|
| 179 |
+
### 다국어 CoT 데이터셋
|
| 180 |
+
|
| 181 |
+
| 이름 | 크기 | 언어 | 설명 | 링크 |
|
| 182 |
+
|------|------|------|------|------|
|
| 183 |
+
| **KAIST Multilingual CoT Collection** | 1.84M CoT | 다국어 | Flan Collection 기반 1060개 태스크. CoT 능력 주입용. | [🤗 kaist-ai/CoT-Collection](https://huggingface.co/datasets/kaist-ai/CoT-Collection) |
|
| 184 |
+
| **OpenO1-SFT** | - | 영어 | O1 스타일 추론 SFT 데이터. 한국어 번역 가능. | [🤗 O1-OPEN/OpenO1-SFT](https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT) |
|
| 185 |
+
| **NuminaMath-TIR** | 860K | 영어 | AI Math Olympiad 수상 데이터. **Tool-Integrated Reasoning**. | [🤗 AI-MO/NuminaMath-TIR](https://huggingface.co/datasets/AI-MO/NuminaMath-TIR) |
|
| 186 |
+
| **NuminaMath-CoT** | 859K | 영어 | Chain-of-Thought 수학 문제 풀이. | [🤗 AI-MO/NuminaMath-CoT](https://huggingface.co/datasets/AI-MO/NuminaMath-CoT) |
|
| 187 |
+
| **OpenMathInstruct-2** | 14M | 영어 | GSM8K/MATH 기반 Llama-3.1-405B 합성 데이터. | [🤗 nvidia/OpenMathInstruct-2](https://huggingface.co/datasets/nvidia/OpenMathInstruct-2) |
|
| 188 |
+
| **AceReason-1.1-SFT** | 4M | 영어 | DeepSeek-R1로 생성한 **수학/코드 추론 SFT**. OpenMathReasoning, OpenCodeReasoning 등 통합. | [🤗 nvidia/AceReason-1.1-SFT](https://huggingface.co/datasets/nvidia/AceReason-1.1-SFT) |
|
| 189 |
+
|
| 190 |
+
### 추론 능력 향상을 위한 모델 (참고)
|
| 191 |
+
|
| 192 |
+
| 모델 | 크기 | 설명 | 링크 |
|
| 193 |
+
|------|------|------|------|
|
| 194 |
+
| **Nemotron-Research-Reasoning-Qwen-1.5B** | 1.5B | ProRL로 학습된 추론 모델. NVIDIA 연구용. | [🤗 nvidia/Nemotron-Research-Reasoning-Qwen-1.5B](https://huggingface.co/nvidia/Nemotron-Research-Reasoning-Qwen-1.5B) |
|
| 195 |
+
| **LLaDA2.0-mini** | 16B | Diffusion LLM. MoE Instruction-tuned. | [🤗 inclusionAI/LLaDA2.0-mini](https://huggingface.co/inclusionAI/LLaDA2.0-mini) |
|
| 196 |
+
| **LLaDA2.0-flash** | 100B | Diffusion LLM. MoE Instruction-tuned. | [🤗 inclusionAI/LLaDA2.0-flash](https://huggingface.co/inclusionAI/LLaDA2.0-flash) |
|
| 197 |
+
|
| 198 |
+
> 💡 **팁**: 영어 CoT 데이터를 한국어로 번역하면 저비용으로 한국어 추론 데이터를 확보할 수 있습니다.
|
| 199 |
+
> 위의 "무료 번역 전략" 섹션을 참고하세요.
|
| 200 |
+
|
| 201 |
+
---
|
| 202 |
+
|
| 203 |
+
## Post-training 데이터셋
|
| 204 |
+
|
| 205 |
+
### SFT (Supervised Fine-Tuning)
|
| 206 |
+
|
| 207 |
+
#### 📌 대규모 통합 데이터셋
|
| 208 |
+
| 이름 | 크기 | 타입 | 설명 | 라이센스 | 링크 |
|
| 209 |
+
|------|------|------|------|----------|------|
|
| 210 |
+
| **KoCommercial-Dataset** | 1.44M | 싱글턴 | 상업적 이용 가능한 데이터 병합. **가장 큰 한국어 SFT 데이터**. | 상업적 가능 | [🤗 MarkrAI/KoCommercial-Dataset](https://huggingface.co/datasets/MarkrAI/KoCommercial-Dataset) |
|
| 211 |
+
| **open-korean-instructions** | 다양 | 혼합 | 공개 한국어 instruction 데이터 통합 저장소. | 다양함 | [🤗 heegyu/open-korean-instructions](https://huggingface.co/datasets/heegyu/open-korean-instructions) |
|
| 212 |
+
| **koVast** | 685K | 멀티턴 | 대규모 멀티턴 한국어 대화 데이터. | - | [🤗 maywell/koVast](https://huggingface.co/datasets/maywell/koVast) |
|
| 213 |
+
| **smol-koreantalk** | 460K | 멀티턴 | SmolLM2 학습 데이터(smol-smoltalk) 한국어 번역. | Apache 2.0 | [🤗 lemon-mint/smol-koreantalk](https://huggingface.co/datasets/lemon-mint/smol-koreantalk) |
|
| 214 |
+
|
| 215 |
+
#### 📌 고품질 번역 데이터셋
|
| 216 |
+
| 이름 | 크기 | 타입 | 설명 | 라이센스 | 링크 |
|
| 217 |
+
|------|------|------|------|----------|------|
|
| 218 |
+
| **ShareGPT DeepL 번역** | 620K(싱글)+84K(멀티) | 멀티턴 | ShareGPT 데이터 DeepL 번역. | CC BY 2.0 KR | [🤗 junelee/sharegpt_deepl_ko](https://huggingface.co/datasets/junelee/sharegpt_deepl_ko) |
|
| 219 |
+
| **KULLM v2** | 153K | 싱글턴 | GPT4ALL, Dolly, Vicuna 데이터 DeepL 번역. | - | [🤗 nlpai-lab/kullm-v2](https://huggingface.co/datasets/nlpai-lab/kullm-v2) |
|
| 220 |
+
| **OpenOrca-gugugo-ko** | 640K+ | 싱글턴 | OpenOrca 한국어 번역 (진행 중). | - | [🤗 squarelike/OpenOrca-gugugo-ko](https://huggingface.co/datasets/squarelike/OpenOrca-gugugo-ko) |
|
| 221 |
+
| **Ko.WizardLM_evol_instruct_V2_196k** | 196K | 싱글턴 | WizardLM evol_instruct 한국어 번역. | - | [🤗 Dataset](https://huggingface.co/datasets/nlp-with-deeplearning/Ko.WizardLM_evol_instruct_V2_196k) |
|
| 222 |
+
|
| 223 |
+
#### 📌 2024-2025 최신 데이터셋 ⭐
|
| 224 |
+
| 이름 | 크기 | 타입 | 설명 | 라이센스 | 링크 |
|
| 225 |
+
|------|------|------|------|----------|------|
|
| 226 |
+
| **Magpie-Pro-MT-300K-ko** | 300K | 멀티턴 | **Magpie 기법**으로 생성된 합성 한국어 instruction 데이터. | - | [🤗 nayohan/Magpie-Pro-MT-300K-v0.1-ko](https://huggingface.co/datasets/nayohan/Magpie-Pro-MT-300K-v0.1-ko) |
|
| 227 |
+
| **KoAlpaca-RealQA** | 18K | 싱글턴 | 2023-2024 ChatKoAlpaca **실제 사용자 대화** 기반. | CC BY-SA 4.0 | [🤗 beomi/KoAlpaca-RealQA](https://huggingface.co/datasets/beomi/KoAlpaca-RealQA) |
|
| 228 |
+
| **Won-Instruct** | 86K | 싱글턴 | **금융 도메인** 특화 한국어 instruction 데이터. KRX 제작. | 확인 필요 | [🤗 KRX-Data/Won-Instruct](https://huggingface.co/datasets/KRX-Data/Won-Instruct) |
|
| 229 |
+
| **ko-limo** | 1K | 싱글턴 | LIMO 논문 데이터 한국어 번역. **추론 능력** 강화용. | - | [🤗 junnei/ko-limo](https://huggingface.co/datasets/junnei/ko-limo) |
|
| 230 |
+
| **ko_llm_annotations v3** | - | 합성 | 한국어 LLM 합성 데이터. 2024년 9월 업데이트. | - | [🤗 devngho/ko_llm_annotations](https://huggingface.co/datasets/devngho/ko_llm_annotations) |
|
| 231 |
+
|
| 232 |
+
#### 📌 도메인 특화 데이터셋
|
| 233 |
+
| 이름 | 크기 | 도메인 | 설명 | 라이센스 | 링크 |
|
| 234 |
+
|------|------|--------|------|----------|------|
|
| 235 |
+
| **HR-Instruct-Math-v0.1** | 30K | 수학 | 한국어 수학 instruction 데이터. | - | [🤗 HAERAE-HUB/HR-Instruct-Math-v0.1](https://huggingface.co/datasets/HAERAE-HUB/HR-Instruct-Math-v0.1) |
|
| 236 |
+
| **orca-math-korean** | 193K | 수학 | Microsoft orca-math 한국어 번역. | - | [🤗 kuotient/orca-math-word-problems-193k-korean](https://huggingface.co/datasets/kuotient/orca-math-word-problems-193k-korean) |
|
| 237 |
+
| **ko_medical_chat** | 3K | 의료 | 의료 대화 데이터. | - | [🤗 squarelike/ko_medical_chat](https://huggingface.co/datasets/squarelike/ko_medical_chat) |
|
| 238 |
+
| **CounselGPT** | 13K+8.7K | 상담 | GPT로 생성한 상담 대화 데이터. | - | [GitHub](https://github.com/MrBananaHuman/CounselGPT) |
|
| 239 |
+
| **glaive-function-calling-v2-ko** | 15.2K | Function Calling | 함수 호출 학습용 데이터. | - | [🤗 heegyu/glaive-function-calling-v2-ko](https://huggingface.co/datasets/heegyu/glaive-function-calling-v2-ko) |
|
| 240 |
+
|
| 241 |
+
---
|
| 242 |
+
|
| 243 |
+
### DPO / Preference 데이터셋
|
| 244 |
+
|
| 245 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 246 |
+
|------|------|------|----------|------|
|
| 247 |
+
| **ko_Ultrafeedback_binarized** | 62K | Ultrafeedback 번역 + 정제. DPO 학습용. | 비상업적* | [🤗 maywell/ko_Ultrafeedback_binarized](https://huggingface.co/datasets/maywell/ko_Ultrafeedback_binarized) |
|
| 248 |
+
| **orca-dpo-pairs-ko** | 36K | 3개 DPO 데이터셋 병합 후 중복 제거. | - | [🤗 SJ-Donald/orca-dpo-pairs-ko](https://huggingface.co/datasets/SJ-Donald/orca-dpo-pairs-ko) |
|
| 249 |
+
| **orca-math-korean-preference** | 193K | 수학 DPO 데이터셋. | - | [🤗 kuotient/orca-math-korean-preference](https://huggingface.co/datasets/kuotient/orca-math-korean-preference) |
|
| 250 |
+
| **K2-Feedback** | 100K | 한국어 평가 능력 향상용. Prometheus 학습 데이터 기반. | - | [🤗 HAERAE-HUB/K2-Feedback](https://huggingface.co/datasets/HAERAE-HUB/K2-Feedback) |
|
| 251 |
+
|
| 252 |
+
> *비상업적: 데이터 직접 상업 사용 불가, 모델 학습 후 상업 사용 가능
|
| 253 |
+
|
| 254 |
+
#### 🔬 OLMo 3 Dolci Post-training Datasets (2025 최신) ⭐
|
| 255 |
+
| 이름 | 크기 | 용도 | 설명 | 링크 |
|
| 256 |
+
|------|------|------|------|------|
|
| 257 |
+
| **Dolci-Think-SFT-7B** | 2.27M | SFT | OLMo 3 7B Think 모델 SFT 데이터. | [🤗 allenai/Dolci-Think-SFT-7B](https://huggingface.co/datasets/allenai/Dolci-Think-SFT-7B) |
|
| 258 |
+
| **Dolci-Think-DPO-7B** | 150K | DPO | OLMo 3 7B Think 모델 DPO 데이터. | [🤗 allenai/Dolci-Think-DPO-7B](https://huggingface.co/datasets/allenai/Dolci-Think-DPO-7B) |
|
| 259 |
+
| **Dolci-Think-RL-7B** | 102K | RL | OLMo 3 7B Think 모델 RL 데이터. | [🤗 allenai/Dolci-Think-RL-7B](https://huggingface.co/datasets/allenai/Dolci-Think-RL-7B) |
|
| 260 |
+
| **Dolci-Instruct-SFT** | 2.15M | SFT | OLMo 3 Instruct 모델 SFT 데이터. | [🤗 allenai/Dolci-Instruct-SFT](https://huggingface.co/datasets/allenai/Dolci-Instruct-SFT) |
|
| 261 |
+
| **Dolci-Instruct-DPO** | 260K | DPO | OLMo 3 Instruct 모델 DPO 데이터. | [🤗 allenai/Dolci-Instruct-DPO](https://huggingface.co/datasets/allenai/Dolci-Instruct-DPO) |
|
| 262 |
+
| **Dolci-Think-SFT-Python** | 1.09M | Code SFT | Python 코드 SFT 믹스. | [🤗 allenai/Dolci-Think-SFT-Python](https://huggingface.co/datasets/allenai/Dolci-Think-SFT-Python) |
|
| 263 |
+
| **Dolci-RL-Zero-Math-7B** | 13.3K | RL Zero | 수학 도메인 RL Zero 데이터. | [🤗 allenai/Dolci-RL-Zero-Math-7B](https://huggingface.co/datasets/allenai/Dolci-RL-Zero-Math-7B) |
|
| 264 |
+
| **Dolci-RL-Zero-Code-7B** | 13.3K | RL Zero | 코드 도메인 RL Zero 데이터. | [🤗 allenai/Dolci-RL-Zero-Code-7B](https://huggingface.co/datasets/allenai/Dolci-RL-Zero-Code-7B) |
|
| 265 |
+
|
| 266 |
+
> 📦 **OLMo 3 Post-training Collection**: [🤗 allenai/Olmo-3-Post-training](https://huggingface.co/collections/allenai/olmo-3-post-training)
|
| 267 |
+
|
| 268 |
+
#### 🚀 NVIDIA Nemotron Post-training v3 Datasets (2025 최신) ⭐
|
| 269 |
+
| 이름 | 크기 | 용도 | 설명 | 링크 |
|
| 270 |
+
|------|------|------|------|------|
|
| 271 |
+
| **Nemotron-Instruction-Following-Chat-v1** | 288K | SFT | Instruction Following Chat 데이터. | [🤗 nvidia/Nemotron-Instruction-Following-Chat-v1](https://huggingface.co/datasets/nvidia/Nemotron-Instruction-Following-Chat-v1) |
|
| 272 |
+
| **Nemotron-Math-Proofs-v1** | 925K | Math | 수학 증명 데이터. | [🤗 nvidia/Nemotron-Math-Proofs-v1](https://huggingface.co/datasets/nvidia/Nemotron-Math-Proofs-v1) |
|
| 273 |
+
| **Nemotron-Math-v2** | - | Math | 수학 Post-training v2. | [🤗 nvidia/Nemotron-Math-v2](https://huggingface.co/datasets/nvidia/Nemotron-Math-v2) |
|
| 274 |
+
| **Nemotron-Science-v1** | 226K | Science | 과학 도메인 데이터. | [🤗 nvidia/Nemotron-Science-v1](https://huggingface.co/datasets/nvidia/Nemotron-Science-v1) |
|
| 275 |
+
| **Nemotron-Agentic-v1** | - | Agentic | 에이전트 학습용 데이터. | [🤗 nvidia/Nemotron-Agentic-v1](https://huggingface.co/datasets/nvidia/Nemotron-Agentic-v1) |
|
| 276 |
+
| **Nemotron-Competitive-Programming-v1** | - | Code | 경쟁 프로그래밍 데이터. | [🤗 nvidia/Nemotron-Competitive-Programming-v1](https://huggingface.co/datasets/nvidia/Nemotron-Competitive-Programming-v1) |
|
| 277 |
+
| **Nemotron-3-Nano-RL-Training-Blend** | - | RL | Nemotron Nano RL 학습 블렌드. | [🤗 nvidia/Nemotron-3-Nano-RL-Training-Blend](https://huggingface.co/datasets/nvidia/Nemotron-3-Nano-RL-Training-Blend) |
|
| 278 |
+
|
| 279 |
+
> 📦 **NVIDIA Nemotron Post-training Collection**: [🤗 nvidia/Nemotron-Post-Training-v3](https://huggingface.co/collections/nvidia/nemotron-post-training-v3)
|
| 280 |
+
|
| 281 |
+
#### 🤖 GRPO / RL 학습용 데이터셋 (DeepSeek-R1 스타일) ⭐
|
| 282 |
+
> **GRPO (Group Relative Policy Optimization)**는 DeepSeek-R1에서 도입된 RL 방법론으로,
|
| 283 |
+
> PPO보다 효율적이며 수학/코드 추론 능력 향상에 탁월합니다.
|
| 284 |
+
|
| 285 |
+
| 이름 | 크기 | 용도 | 설명 | 링크 |
|
| 286 |
+
|------|------|------|------|------|
|
| 287 |
+
| **NuminaMath-TIR** | 860K | Math GRPO | AI Math Olympiad 수상 데이터. **Tool-Integrated Reasoning**. | [🤗 AI-MO/NuminaMath-TIR](https://huggingface.co/datasets/AI-MO/NuminaMath-TIR) |
|
| 288 |
+
| **NuminaMath-CoT** | 859K | Math GRPO | Chain-of-Thought 수학 문제 풀이. | [🤗 AI-MO/NuminaMath-CoT](https://huggingface.co/datasets/AI-MO/NuminaMath-CoT) |
|
| 289 |
+
| **OpenMathInstruct-2** | 14M | Math | GSM8K/MATH 기반 Llama-3.1-405B 합성 데이터. | [🤗 nvidia/OpenMathInstruct-2](https://huggingface.co/datasets/nvidia/OpenMathInstruct-2) |
|
| 290 |
+
| **Dolci-RL-Zero-Math-7B** | 13.3K | GRPO | OLMo 3 수학 도메인 RL Zero 데이터. | [🤗 allenai/Dolci-RL-Zero-Math-7B](https://huggingface.co/datasets/allenai/Dolci-RL-Zero-Math-7B) |
|
| 291 |
+
| **Dolci-RL-Zero-Code-7B** | 13.3K | GRPO | OLMo 3 코드 도메인 RL Zero 데이터. | [🤗 allenai/Dolci-RL-Zero-Code-7B](https://huggingface.co/datasets/allenai/Dolci-RL-Zero-Code-7B) |
|
| 292 |
+
| **Nemotron-3-Nano-RL-Training-Blend** | - | GRPO | Nemotron Nano RL 학습 블렌드. | [🤗 nvidia/Nemotron-3-Nano-RL-Training-Blend](https://huggingface.co/datasets/nvidia/Nemotron-3-Nano-RL-Training-Blend) |
|
| 293 |
+
|
| 294 |
+
> 📚 **GRPO 구현**: HuggingFace TRL 라이브러리의 `GRPOTrainer` 클래스 사용 [📖 TRL GRPO 문서](https://huggingface.co/docs/trl/main/en/grpo_trainer)
|
| 295 |
+
|
| 296 |
+
#### 🇰🇷 한국어 수학 추론 데이터셋
|
| 297 |
+
> 💡 위 [한국어 추론 데이터셋](#한국어-추론-데이터셋-) 및 [도메인 특화 데이터셋](#-도메인-특화-데이터셋)의 **NuminaMath-CoT-Ko**, **orca-math-korean** 참조
|
| 298 |
+
|
| 299 |
+
---
|
| 300 |
+
|
| 301 |
+
### RLHF / RM 데이터셋
|
| 302 |
+
|
| 303 |
+
| 이름 | 크기 | 설명 | 라이센스 | 링크 |
|
| 304 |
+
|------|------|------|----------|------|
|
| 305 |
+
| **ko_hh-rlhf-20k_filtered** | 20K | Anthropic hh-rlhf 한국어 번역 (필터링). | - | [🤗 maywell/ko_hh-rlhf-20k_filtered](https://huggingface.co/datasets/maywell/ko_hh-rlhf-20k_filtered) |
|
| 306 |
+
| **hh-rlhf-ko** | 113K | Anthropic hh-rlhf 전체 번역. | - | [🤗 heegyu/hh-rlhf-ko](https://huggingface.co/datasets/heegyu/hh-rlhf-ko) |
|
| 307 |
+
| **PKU-SafeRLHF-ko** | 164K | PKU 안전 RLHF 데이터 번역. | - | [🤗 heegyu/PKU-SafeRLHF-ko](https://huggingface.co/datasets/heegyu/PKU-SafeRLHF-ko) |
|
| 308 |
+
| **kor_ethical_question_answer** | 29.1K | AI 윤리적/비윤리적 QA 데이터. | - | [🤗 MrBananaHuman/kor_ethical_question_answer](https://huggingface.co/datasets/MrBananaHuman/kor_ethical_question_answer) |
|
| 309 |
+
| **korean_rlhf_dataset** | 107K | 성균관대 산학협력 SFT 데이터. | - | [🤗 jojo0217/korean_rlhf_dataset](https://huggingface.co/datasets/jojo0217/korean_rlhf_dataset) |
|
| 310 |
+
| **AIHub RLHF Dataset** | SFT 13K, RM 33K, PPO 33K | 공식 AIHub 제공. RM 데이터는 5개 답변 순위 포함. | - | [AI Hub](https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=data&dataSetSn=71748) |
|
| 311 |
+
|
| 312 |
+
---
|
| 313 |
+
|
| 314 |
+
## 무료 번역 전략 (영어 - 한국어)
|
| 315 |
+
> **핵심 아이디어**: 영어 고품질 데이터셋은 풍부하므로, 무료 번역 도구를 활용하여 한국어 데이터를 확보합니다.
|
| 316 |
+
>
|
| 317 |
+
> **비용 절감**: 상용 번역 API 대신 무료 도구를 활용하면 대규모 데이터셋도 무료로 구축 가능합니다.
|
| 318 |
+
|
| 319 |
+
### 무료 번역 도구 비교
|
| 320 |
+
|
| 321 |
+
| 도구 | 무료 한도 | 한국어 품질 | 특징 | 설치/사용법 |
|
| 322 |
+
|------|-----------|-------------|------|-------------|
|
| 323 |
+
| **Google Translate (비공식)** | 무제한 | ⭐⭐⭐⭐⭐ | 가장 높은 한국어 품질, 비공식 라이브러리 | `pip install googletrans==4.0.0-rc1` |
|
| 324 |
+
| **DeepL API Free** | 500K chars/month | ⭐⭐⭐⭐ | 유럽어 최고, 한국어도 양호 | [API 키 신청](https://www.deepl.com/pro-api) |
|
| 325 |
+
| **LibreTranslate** | 무제한 (셀프호스팅) | ⭐⭐⭐ | 오픈소스, 로컬 실행 가능 | `pip install libretranslate` |
|
| 326 |
+
| **MarianMT (HuggingFace)** | 무제한 | ⭐⭐⭐ | 오픈소스 NMT 모델, 완전 로컬 | `transformers` 라이브러리 |
|
| 327 |
+
| **NLLB (Meta)** | 무제한 | ⭐⭐⭐ | 200개 언어, 고품질 다국어 번역 | [🤗 facebook/nllb-200-distilled-600M](https://huggingface.co/facebook/nllb-200-distilled-600M) |
|
| 328 |
+
| **lfm2-1.2b-koen-mt-v8-rl-10k-merged-GGUF** | 무제한 | ⭐⭐⭐⭐ | 1.2B 수준에서 최고 성능을 보이는 한국어-영어 번역 모델 | [🤗 gyung/lfm2-1.2b-koen-mt-v8-rl-10k-merged-GGUF](https://huggingface.co/gyung/lfm2-1.2b-koen-mt-v8-rl-10k-merged-GGUF) |
|
| 329 |
+
|
| 330 |
+
|
| 331 |
+
### 추천 번역 파이프라인
|
| 332 |
+
|
| 333 |
+
```
|
| 334 |
+
┌─────────────────────────────────────────────────────────────────────┐
|
| 335 |
+
│ 1. 영어 데이터셋 선택 (예: OpenOrca, Alpaca, WizardLM) │
|
| 336 |
+
├─────────────────────────────────────────────────────────────────────┤
|
| 337 |
+
│ 2. Google Translate (비공식) 또는 LFM2로 1차 번역 │
|
| 338 |
+
│ → 무료이며 품질이 가장 좋음 │
|
| 339 |
+
├─────────────────────────────────────────────────────────────────────┤
|
| 340 |
+
│ 3. 품질 필터링 (선택사항) │
|
| 341 |
+
│ → LLM으로 번역 품질 평가 또는 rule-based 필터링 │
|
| 342 |
+
├─────────────────────────────────────────────────────────────────────┤
|
| 343 |
+
│ 4. 최종 한국어 데이터셋 생성 │
|
| 344 |
+
└─────────────────────────────────────────────────────────────────────┘
|
| 345 |
+
```
|
| 346 |
+
|
| 347 |
+
### Google Translate 사용 예시 (Python)
|
| 348 |
+
|
| 349 |
+
```python
|
| 350 |
+
from googletrans import Translator
|
| 351 |
+
|
| 352 |
+
translator = Translator()
|
| 353 |
+
|
| 354 |
+
def translate_to_korean(text):
|
| 355 |
+
try:
|
| 356 |
+
result = translator.translate(text, src='en', dest='ko')
|
| 357 |
+
return result.text
|
| 358 |
+
except Exception as e:
|
| 359 |
+
return None
|
| 360 |
+
|
| 361 |
+
# 대량 번역 시 rate limiting 주의
|
| 362 |
+
# 멀티스레딩 + 재시도 로직 권장
|
| 363 |
+
```
|
| 364 |
+
|
| 365 |
+
> 💡 **팁**: 대규모 번역 시 [Argilla Distilabel](https://github.com/argilla-io/distilabel) 또는 [Curator](https://github.com/bespokelabsai/curator/) 같은 도구를 활용하면 멀티스레딩, 자동 재시도, 체크포인트 등을 지원합니다.
|
| 366 |
+
|
| 367 |
+
---
|
| 368 |
+
|
| 369 |
+
## 평가용 데이터셋
|
| 370 |
+
|
| 371 |
+
| 이름 | 크기 | 타입 | 설명 | 링크 |
|
| 372 |
+
|------|------|------|------|------|
|
| 373 |
+
| **KMMLU** | 243K | MCQA | 45개 주제 전문가 수준 한국어 벤치마크. | [🤗 HAERAE-HUB/KMMLU](https://huggingface.co/datasets/HAERAE-HUB/KMMLU) |
|
| 374 |
+
| **HAE-RAE-BENCH** | 1.5K | MCQA | 어휘, 역사, 상식, 독해 평가. | [GitHub](https://github.com/HAETAE-project/HAE-RAE-BENCH) |
|
| 375 |
+
| **CSAT-QA** | 0.9K | MCQA | 국어 수능 문제. | [🤗 HAERAE-HUB/CSAT-QA](https://huggingface.co/datasets/HAERAE-HUB/CSAT-QA) |
|
| 376 |
+
| **K2-Eval** | 90 | 생성 | 한국 문화 지식 필요한 90개 지시문. GPT-4 평가. | [🤗 HAERAE-HUB/K2-Eval](https://huggingface.co/datasets/HAERAE-HUB/K2-Eval) |
|
| 377 |
+
| **KorMedMCQA** | <1K | MCQA | 한국어 의료 QA 벤치마크. | [🤗 sean0042/KorMedMCQA](https://huggingface.co/datasets/sean0042/KorMedMCQA) |
|
| 378 |
+
| **LogicKor** | - | 다분야 | 한국어 사고력 벤치마크. | [🤗 Leaderboard](https://huggingface.co/spaces/instructkr/LogicKor-leaderboard) |
|
| 379 |
+
|
| 380 |
+
---
|
| 381 |
+
|
| 382 |
+
## 유용한 컬렉션
|
| 383 |
+
|
| 384 |
+
| 컬렉션 | 설명 | 링크 |
|
| 385 |
+
|--------|------|------|
|
| 386 |
+
| **나요한님 번역 데이터** | 영어 데이터셋 한국어 번역. llama3-instrucTrans 사용. | [🤗 Collection](https://huggingface.co/collections/nayohan/translated-en-ko-dataset-6665023b1036d124ede5f81c) |
|
| 387 |
+
| **나요한님 Magpie 번역** | Magpie 데이터셋 한국어 번역. | [🤗 Collection](https://huggingface.co/collections/youjunhyeok/magpie-ko-66cbc570a9891d5b43a170d9) |
|
| 388 |
+
| **유준혁님 번역 데이터** | 영한 번역 데이터셋 모음. | [🤗 Collection](https://huggingface.co/collections/youjunhyeok/en-ko-translate-6703474b419fcb9e5d6a7852) |
|
| 389 |
+
| **송영숙님 Korean Dataset** | 허깅페이스 한국어 데이터셋 정리 (2024.10 기준). | [GitHub](https://github.com/songys/huggingface_KoreanDataset) |
|
| 390 |
+
|
| 391 |
+
---
|
| 392 |
+
|
| 393 |
+
## 참고 자료
|
| 394 |
+
|
| 395 |
+
### 합성 데이터 구축
|
| 396 |
+
- [ko-genstruct](https://github.com/iKnowLab-Projects/ko-genstruct) - 한국어 합성 데이터 생성
|
| 397 |
+
- [evolve-instruct](https://github.com/lcw99/evolve-instruct) - Instruction 증강 기법
|
| 398 |
+
|
| 399 |
+
### 평가 플랫폼
|
| 400 |
+
- [Ko Chatbot Arena](https://huggingface.co/spaces/instructkr/ko-chatbot-arena-leaderboard) - 한국어 챗봇 ELO 랭킹
|
| 401 |
+
- [LogicKor Leaderboard](https://huggingface.co/spaces/instructkr/LogicKor-leaderboard) - 다분야 사고력 평가
|
| 402 |
+
- [호랑이 LLM 리더보드](https://wandb.ai/wandb-korea/korean-llm-leaderboard/reports) - W&B 한국어 LLM 평가
|
| 403 |
+
|
| 404 |
+
|
| 405 |
+
### 🇰🇷 한국 기업 LLM 기술 보고서 (데이터 전략 참고)
|
| 406 |
+
| 기업 | 모델 | 핵심 전략 | 보고서 |
|
| 407 |
+
|------|------|----------|--------|
|
| 408 |
+
| **Upstage** | Solar Open | 4.5T 합성 데이터 + Progressive Curriculum + SnapPO | [📜 Technical Report](https://huggingface.co/upstage/Solar-Open-100B/blob/main/solar-open-technical-report.pdf) |
|
| 409 |
+
| **LG AI Research** | K-EXAONE | 6개 국어 + 256K Context + MoE 구조 | [📜 arXiv](https://arxiv.org/pdf/2601.01739) |
|
| 410 |
+
| **SK Telecom** | A.X K1 | 10T 토큰 + Multi-stage Pipeline + Think-Fusion | [📜 Tech Report](https://github.com/SKT-AI/A.X-K1/releases/download/v1.0/A_X_Tech_Report.pdf) |
|
| 411 |
+
|
| 412 |
+
---
|
| 413 |
+
|
| 414 |
+
## 🎯 Yaongi 프로젝트 권장 데이터 파이프라인
|
| 415 |
+
|
| 416 |
+
> ⚠️ **핵심 인사이트** (Solar Open, K-EXAONE, A.X K1 기술 보고서 기반):
|
| 417 |
+
> - 단순 웹 크롤링만으로는 부족 → **합성 데이터(Synthetic Data)** 필수
|
| 418 |
+
> - **커리큘럼 학습** (Progressive Curriculum): 단계별 데이터 품질 조절
|
| 419 |
+
> - 500M 모델은 용량이 작으므로 **압축적이고 밀도 높은 데이터** 필요
|
| 420 |
+
|
| 421 |
+
### Phase 1: Pre-training (500M 모델, 300B 토큰)
|
| 422 |
+
|
| 423 |
+
```
|
| 424 |
+
┌─────────────────────────────────────────────────────────────┐
|
| 425 |
+
│ 영어 (50% = 150B) 한국어 (50% = 150B) │
|
| 426 |
+
├──────────────────────────────────��──────────────────────────┤
|
| 427 |
+
│ • FineWeb-Edu • Korean Wikipedia │
|
| 428 |
+
│ • SmolLM-Corpus • korean_textbooks (합성) │
|
| 429 |
+
│ • Nemotron-CC • aihub-en-ko-translation │
|
| 430 |
+
│ • The Stack (코드) • KOREAN-SyntheticText │
|
| 431 |
+
└─────────────────────────────────────────────────────────────┘
|
| 432 |
+
```
|
| 433 |
+
|
| 434 |
+
**한국어 Pre-training 데이터 확보 전략:**
|
| 435 |
+
- 허깅페이스에 있는 합성 데이터셋(korean_textbooks, KOREAN-SyntheticText) 활용
|
| 436 |
+
- 한영 번역 말뭉치(aihub-en-ko-translation-12m) Pre-training에 포함
|
| 437 |
+
- 부족 시 Post-training 데이터(KoCommercial, koVast) 일부 Pre-training에 활용
|
| 438 |
+
|
| 439 |
+
#### 📊 커리큘럼 학습 전략 (Solar Open 참조)
|
| 440 |
+
| 단계 | 토큰 | 데이터 구성 | 목표 |
|
| 441 |
+
|------|------|-------------|------|
|
| 442 |
+
| **Phase 1a** | 0~200B | 일반 한국어/영어/코드 혼합 | 기초 언어 능력 |
|
| 443 |
+
| **Phase 1b** | 200~280B | 고품질 교과서 + 전문 텍스트 | 지식 밀도 |
|
| 444 |
+
| **Phase 1c (Annealing)** | 280~300B | **합성 CoT 데이터 집중** | 추론 능력 극대화 |
|
| 445 |
+
|
| 446 |
+
### Phase 2: Mid-training / Continued Pre-training
|
| 447 |
+
|
| 448 |
+
```
|
| 449 |
+
┌─────────────────────────────────────────────────────────────┐
|
| 450 |
+
│ 고품질 한국어 집중 (50~100B 토큰) │
|
| 451 |
+
├─────────────────────────────────────────────────────────────┤
|
| 452 |
+
│ • Dolma3 Dolmino Mix (OLMo 3 스타일) │
|
| 453 |
+
│ • Korean Pretraining Collection │
|
| 454 |
+
│ • 뉴스 기사 + 사설 (논리적 글쓰기) │
|
| 455 |
+
│ • 합성 한국어 CoT 데이터 (GPT-4/Claude로 생성) │
|
| 456 |
+
└─────────────────────────────────────────────────────────────┘
|
| 457 |
+
```
|
| 458 |
+
|
| 459 |
+
### Phase 3: Post-training (SFT)
|
| 460 |
+
|
| 461 |
+
```
|
| 462 |
+
┌─────────────────────────────────────────────────────────────┐
|
| 463 |
+
│ 1순위: KoCommercial-Dataset (1.44M) - 상업적 이용 가능 │
|
| 464 |
+
│ 2순위: open-korean-instructions 통합 데이터 │
|
| 465 |
+
│ 3순위: Magpie-Pro-MT-300K-ko (합성 데이터) │
|
| 466 |
+
│ ───────────────────────────────────────────────────────────│
|
| 467 |
+
│ 💡 English 참고: Dolci-Instruct-SFT, Nemotron-IF-Chat │
|
| 468 |
+
└─────────────────────────────────────────────────────────────┘
|
| 469 |
+
```
|
| 470 |
+
|
| 471 |
+
### Phase 4: Alignment (DPO/RLHF → **GRPO**)
|
| 472 |
+
|
| 473 |
+
> ⭐ **GRPO (Group Relative Policy Optimization)** 기반 RL이 핵심!
|
| 474 |
+
> DeepSeek-R1에서 입증된 방법으로, PPO보다 효율적이며 수학/코드 추론에 탁월합니다.
|
| 475 |
+
|
| 476 |
+
```
|
| 477 |
+
┌─────────────────────────────────────────────────────────────┐
|
| 478 |
+
│ Stage 1 - DPO (기본 정렬) │
|
| 479 |
+
│ • ko_Ultrafeedback_binarized + orca-dpo-pairs-ko │
|
| 480 |
+
├─────────────────────────────────────────────────────────────┤
|
| 481 |
+
│ Stage 2 - GRPO (추론 강화) ⭐ │
|
| 482 |
+
│ • NuminaMath-CoT-Ko (수학 추론) │
|
| 483 |
+
│ • NuminaMath-TIR (Tool-Integrated Reasoning) │
|
| 484 |
+
│ • Dolci-RL-Zero-Math, Dolci-RL-Zero-Code │
|
| 485 |
+
└─────────────────────────────────────────────────────────────┘
|
| 486 |
+
```
|
| 487 |
+
|
| 488 |
+
### 💡 합성 데이터 활용 가이드
|
| 489 |
+
|
| 490 |
+
> 500M 모델은 허깅페이스에 있는 **기존 합성 데이터셋**을 활용하면 됩니다.
|
| 491 |
+
> 직접 생성할 필요 없이 아래 데이터셋들을 바로 사용하세요!
|
| 492 |
+
|
| 493 |
+
| 카테고리 | 추천 데이터셋 | 용량 | 효과 |
|
| 494 |
+
|----------|---------------|------|------|
|
| 495 |
+
| **한국어 교과서** | maywell/korean_textbooks | 1~10M | 지식 밀도 ↑ |
|
| 496 |
+
| **한국어 합성** | KOREAN-SyntheticText-1.5B | 1.5B | Pre-training 확장 |
|
| 497 |
+
| **한영 번역** | aihub-en-ko-translation-12m | 12M | 지식 주입 |
|
| 498 |
+
| **��학 추론** | NuminaMath-CoT-Ko, orca-math-korean | 200K+ | 추론 능력 ↑ |
|
| 499 |
+
| **멀티턴 대화** | Magpie-Pro-MT-300K-ko | 300K | SFT 품질 ↑ |
|
| 500 |
+
|
| 501 |
+
---
|
| 502 |
+
|
| 503 |
+
## 참고 논문
|
| 504 |
+
|
| 505 |
+
> 아래 논문들에서 LLM 학습 전략, 데이터셋 구성, RL 기법 등의 인사이트를 얻을 수 있습니다.
|
| 506 |
+
|
| 507 |
+
### RL 학습 및 추론 능력 향상
|
| 508 |
+
|
| 509 |
+
| 논문 | 핵심 기여 | 관련 리소스 | 링크 |
|
| 510 |
+
|------|----------|-------------|------|
|
| 511 |
+
| **ProRL: Prolonged RL Expands Reasoning Boundaries** | 장기간 RL로 base 모델에서 불가능한 추론 전략 발견. KL divergence 제어, reference policy resetting. | [🤗 Nemotron-Research-Reasoning-Qwen-1.5B](https://huggingface.co/nvidia/Nemotron-Research-Reasoning-Qwen-1.5B) | [📜 arXiv:2505.24864](https://arxiv.org/abs/2505.24864) |
|
| 512 |
+
| **Stabilizing RL with LLMs** | 30B MoE 모델 RL 안정화 레시피. Importance sampling, Clipping, **Routing Replay** (MoE 전용). | - | [📜 arXiv:2512.01374](https://arxiv.org/abs/2512.01374) |
|
| 513 |
+
|
| 514 |
+
### Agent 및 Deep Research
|
| 515 |
+
|
| 516 |
+
| 논문 | 핵심 기여 | 관련 리소스 | 링크 |
|
| 517 |
+
|------|----------|-------------|------|
|
| 518 |
+
| **Step-DeepResearch** | Atomic Capability 기반 합성 데이터 생성. Progressive Training (Mid-training → SFT → RL). 32B 모델로 O1급 성능. | [💻 GitHub](https://github.com/stepfun-ai/StepDeepResearch), ADR-Bench (중국어 벤치마크) | [📜 arXiv:2512.20491](https://arxiv.org/abs/2512.20491) |
|
| 519 |
+
|
| 520 |
+
### 모델 아키텍처 및 학습 기법
|
| 521 |
+
|
| 522 |
+
| 논문 | 핵심 기여 | 관련 리소스 | 링크 |
|
| 523 |
+
|------|----------|-------------|------|
|
| 524 |
+
| **LLaDA 2.0: Scaling Diffusion LLM to 100B** | AR → Diffusion LLM 변환. 3-phase Block-level WSD 학습. Parallel decoding으로 효율적 추론. | [🤗 LLaDA 2.0 Collection](https://huggingface.co/collections/inclusionAI/llada-20), [💻 dFactory](https://github.com/inclusionAI/dFactory), [💻 dInfer](https://github.com/inclusionAI/dInfer) | [📜 arXiv:2512.15745](https://arxiv.org/abs/2512.15745) |
|
| 525 |
+
| **Code Foundation Models to Agents** | 코드 LLM 전체 생명주기 서베이. Scaling law, 데이터 구성, RL 실험. | 코드 Pre-training, SFT, RL 실험 데이터 | [📜 arXiv:2511.18538](https://arxiv.org/abs/2511.18538) |
|
| 526 |
+
|
| 527 |
+
### 논문에서 배울 수 있는 핵심 인사이트
|
| 528 |
+
|
| 529 |
+
1. **ProRL**: 장기간 RL 학습이 base 모델에서 접근 불가능한 추론 전략을 발견할 수 있음
|
| 530 |
+
2. **Step-DeepResearch**: 복잡한 태스크를 **원자적 능력(Atomic Capabilities)**으로 분해하여 학습
|
| 531 |
+
3. **Stabilizing RL**: MoE 모델에서 **Routing Replay**가 정책 staleness 완화에 필수적
|
| 532 |
+
4. **LLaDA 2.0**: Diffusion LLM이 AR 모델과 경쟁 가능하며, parallel decoding으로 추론 효율화
|
| 533 |
+
|
| 534 |
+
---
|
| 535 |
+
|
| 536 |
+
## 📖 외부 참고 자료
|
| 537 |
+
|
| 538 |
+
### 데이터셋 큐레이션
|
| 539 |
+
- [mlabonne/llm-datasets](https://github.com/mlabonne/llm-datasets) - Post-training용 데이터셋 및 도구 큐레이션 리스트 ⭐
|
| 540 |
+
- [open-korean-instructions](https://github.com/HeegyuKim/open-korean-instructions) - 이 README의 주요 참고 자료
|
| 541 |
+
|
| 542 |
+
### 데이터 도구
|
| 543 |
+
- [Curator](https://github.com/bespokelabsai/curator/) - 합성 데이터 생성 파이프라인
|
| 544 |
+
- [Distilabel](https://github.com/argilla-io/distilabel) - SFT/DPO 데이터 생성 및 증강
|
| 545 |
+
- [Argilla](https://argilla.io/) - 데이터 필터링 및 어노테이션 플랫폼
|
| 546 |
+
|
| 547 |
+
---
|
| 548 |
+
|
| 549 |
+
> 📅 **Last Updated**: 2026-01-12
|
| 550 |
+
>
|
| 551 |
+
> 💡 **기여하기**: 새로운 데이터셋 발견 시 PR 또는 Issue로 알려주세요!
|