LLM-jp-4.1 Post-Training Dataset
Collection
4 items • Updated • 2
ID stringlengths 29 33 | messages stringlengths 455 16.4k | chosen_analysis stringlengths 11 12.7k | chosen_final stringlengths 1 21.1k | rejected_analysis stringlengths 11 8.17k | rejected_final stringlengths 1 14.2k |
|---|---|---|---|---|---|
daring_anteater_en-28265-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2029-06-10", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need to provide code? Probably give HTML/JS example. Simple. Provide minimal. | **HTML + CSS + JavaScript (single file)**
```html
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Simple Platformer</title>
<style>
body { margin:0; overflow:hidden; background:#222; }
canvas { display:block; margin:auto; background:#87CEEB; }
</style>
</head>
<body>
<canvas id="game"><... | We need to provide code? Probably give HTML/JS example. Simple. | **HTML (index.html)**
```html
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Simple Platformer</title>
<style>
body { margin:0; overflow:hidden; background:#222; }
canvas { display:block; margin:auto; background:#87CEEB; }
.debug { position:absolute; color:#f... |
daring_anteater_en-66583-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2027-01-21", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need answer concise. | The KION Group recognizes **personnel provisions** covering partial‑retirement obligations, long‑service awards, annual bonuses and severance pay.
These obligations are measured **on an individual contractual basis** (i.e., each employee’s specific contract terms determine the amount). | Need answer succinctly. | The KION Group recognizes **personnel provisions** such as partial‑retirement obligations, long‑service awards, annual bonuses and severance pay.
These obligations are measured based on each employee’s **individual contractual arrangement** (i.e., the specific terms agreed with the employee). |
daring_anteater_en-31413-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2026-01-21", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need list examples and comparison. Provide concise answer. | **Gift‑bringing figures in other European regions & their festivals vs. Italy’s Epiphany (Befana)**
| Region / Country | Figure (local name) | Time of year & main festival | What the figure brings | Key similarities with the Italian Befana/Epiphany |
|------------------|---------------------|------------------------... | Need list examples, brief comparison. | **Gift‑bringing figures outside Italy and how their festivals line up with Italy’s Epiphany (6 January)**
| Region / Country | Figure (local name) | Main Festival & Date | What the figure brings | Key points of comparison with the Italian *Befana* |
|------------------|----------------------|----------------------|-... |
daring_anteater_en-74679-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2026-07-23", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need produce ~500 word paper with citations. | **The Italian and Northern Renaissances: A Comparative Analysis**
The Renaissance, spanning roughly the fourteenth through sixteenth centuries, manifested differently across Europe. While both regions shared an overarching humanist ethos and renewed interest in classical antiquity, their developmental trajectories, ex... | We need produce <=500 words academic paper with citations. Provide comparison. Use references maybe fictional but plausible. | **The Italian and Northern Renaissances: A Comparative Analysis**
Both the Italian and Northern Renaissances represent pivotal moments in European cultural history, yet they evolved through distinct socio‑political contexts, aesthetic priorities, and intellectual currents. While sharing core Humanist ideals—revival of... |
daring_anteater_en-63775-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2029-10-18", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need to provide modified function and brief explanation. Follow guidelines: answer only what's asked. Provide code and explanation. | **Modified function**
```python
def bin_to_signed_int(binary_str: str) -> int:
"""Convert a binary string (two’s‑complement) to a signed integer.
Args:
binary_str: Binary digits possibly prefixed with ``0b``.
May include a leading sign bit (the most‑significant bit).
Returns:
... | We need to provide modified function handling signed two's complement. Provide explanation of sign bit interpretation. Follow instructions: answer only what's asked. So give code and brief explanation. | ```python
def bin_to_int(binary_str: str) -> int:
"""
Convert a binary string to a signed integer using two's‑complement notation.
Supports an optional ``0b``/``0B`` prefix and an optional leading sign bit
(e.g., ``-0b1010``). The length of the remaining bits determines the word size;
the most‑sign... |
daring_anteater_en-3742-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2028-03-26", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need to give explanation arrangement and randomization steps. Concise but enough. | **Treatment‑doctor assignment (Latin‑square layout)**
| | Day 1 | Day 2 | Day 3 | Day 4 |
|----------|-------|-------|-------|-------|
| **Doctor A** | T₁ | T₂ | T₃ | T₄ |
| **Doctor B** | T₂ | T₃ | T₄ | T₁ |
| **Doctor C** | T₃ | T₄ | T₁ | T₂ |
| **Doctor D** | T₄ | T₁ | T₂ | ... | Need to explain arrangement and randomization steps. Provide concise explanation. | **Treatment‑doctor assignment (Latin‑square layout)**
| Run | Doctor 1 | Doctor 2 | Doctor 3 | Doctor 4 |
|-----|----------|----------|----------|----------|
| 1 | A | B | C | D |
| 2 | B | C | D | A |
| 3 | C | D | A | B |
| 4... |
daring_anteater_en-75113-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2030-10-09", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need to provide brief outline steps and key deadlines. Provide concise list. | **Activating the Internship – Step‑by‑Step Outline & Key Deadlines**
1. **Confirm Eligibility (≤ 30 days before start)**
- Verify graduation status, required credits, and any prerequisite courses.
- Submit proof of eligibility (transcript/diploma) via the portal.
2. **Apply through the Employer Portal (deadline... | We need to provide brief outline steps and key deadlines. No extra fluff. | **Activation Steps & Key Deadlines**
1. **Submit Application Materials**
- *What*: CV, cover letter, transcript, and reference letters.
- *Deadline*: 15 May (application window closes).
2. **Complete Online Assessment**
- *What*: Aptitude test + situational judgment questionnaire.
- *Deadline*: 22... |
daring_anteater_en-19085-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2028-07-10", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need concise explanation. | **Relation to the logistic‑loss derivation**
The binary cross‑entropy (logistic) loss for one example \((x^{(i)},y^{(i)})\) is
\[
\ell^{(i)}(\theta)= -\Big[ y^{(i)}\log h_\theta(x^{(i)})
+ (1-y^{(i)})\log\bigl(1-h_\theta(x^{(i)})\bigr)\Big],
\qquad
h_\theta(x)=\sigma(\theta^\top x)=\frac{1}{1+e^{-\theta^\to... | We need concise explanation, no extra. | **Relation to the logistic‐loss derivative**
The logistic‑regression objective is
\[
J(\theta)=\frac{1}{m}\sum_{i=1}^{m}
\Big[-y^{(i)}\log \sigma(z^{(i)})-(1-y^{(i)})\log(1-\sigma(z^{(i)}))\Big],
\qquad
z^{(i)}=x^{(i)\top}\theta,\;\;
\sigma(t)=\frac{1}{1+e^{-t}} .
\]
For each sample
\[
\frac{\partial}{\partial... |
daring_anteater_en-57673-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2026-06-30", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | Need to explain. Provide calculus derivative and AM-GM. | Because the hypotenuse is fixed at the diameter \(c=100\), the legs \(a\) and \(b\) must satisfy
\[
a^{2}+b^{2}=c^{2}=10\,000 .\tag{1}
\]
The area we want to maximize is
\[
A=\tfrac12 ab .
\]
So it suffices to find the largest possible value of the product \(P=ab\) under the constraint (1).
---
### 1. Calculu... | We need to explain. Provide concise explanation using AM-GM or calculus. | Because the hypotenuse is fixed at the diameter, the legs \(a\) and \(b\) must satisfy
\[
a^{2}+b^{2}=R^{2}\qquad(R=50).
\]
The area of the right‑triangle is
\[
A=\tfrac12 ab .
\]
So maximizing the area is equivalent to maximizing the product \(ab\) under the constraint \(a^{2}+b^{2}=R^{2}\).
---
### Method 1... |
daring_anteater_en-9761-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2027-10-16", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need explain steps. | **Step 1 – Write each number with its digits labeled**
| Symbol | Value (base 10) |
|--------|----------------|
| \(120_4\) | \(1·4^2 + 2·4^1 + 0·4^0 = 16 + 8 + 0 = 24\) |
| \(13_4\) | \(1·4^1 + 3·4^0 = 4 + 3 = 7\) |
| \(2_4\) | \(2·4^0 = 2\) |
So
\(120_4 = 24_{10},\; 13_4 = 7_{10},\; 2_4 = 2_{10}\).
---
**S... | We need explain steps. Provide conversion decimal etc. | **Step 1 – Convert each base‑4 number to decimal**
| Base‑4 | Calculation (powers of 4) | Decimal value |
|--------|--------------------------|---------------|
| \(120_4\) | \(1·4^2 + 2·4^1 + 0·4^0 = 1·16 + 2·4 + 0 = 20\) | 20 |
| \(13_4\) | \(1·4^1 + 3·4^0 = 1·4 + 3 = 7\) | 7 |
| \(2_4\) | \(2·4^0 = 2\... |
daring_anteater_en-96902-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2028-07-26", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | Provide list 5 with descriptions concise. | **1. Joseph E. Engelberger – “Father of Industrial Robotics”**
*Achievements*: Co‑developed the first commercially successful industrial robot, Unimate (1961), and founded the world’s first robot company, Unimation Inc. His work established the practical use of programmable automation in manufacturing, setting the te... | Provide list 5 with descriptions. | **1. Joseph E. Stilgoe (1938‑2017)**
*Pioneered modern autonomous mobile robots.* He developed the first practical “self‑navigating” robot—the Stanford Cart—in the early 1970s, introducing real‑time obstacle avoidance using stereo vision and closed‑loop control. His work laid the foundation for today’s autonomous veh... |
daring_anteater_en-85561-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2028-11-18", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | Need tweet <=140 chars summarizing discount. | Follow us on IG? Mention it when you drop off your PC/iPhone & get 17% off—Richmond & Mechanicsville, week of Mar 28! | Need tweet <=140 chars. | Follow @ALBTech on Instagram, mention it when you drop off your PC/iPhone repair, and snag a 17% discount—week of Mar 28! Richmond & Mechanicsville. |
daring_anteater_en-21214-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2027-09-05", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need to give step-by-step derivation and assumptions. Provide detailed explanation. | ### 1. Deriva pas a pas l’expressió de \(u^{*}\)
Suposem un problema de control continu
\[
\begin{aligned}
&\dot x = f(x,u), &&x\in\mathbb R^{n},\; u\in\mathcal U\subset\mathbb R^{m}\\[2mm]
&J=\int_{0}^{\infty} \ell(x(t),u(t))\,dt \qquad (\text{cost integral})
\end{aligned}
\]
i que la funció de valor potencial és... | We need to give step-by-step derivation and explanation of assumptions. Provide concise but detailed answer. | ### Pas a pas: derivació de \(u^{\ast}\) a partir de
\[
\frac{\partial}{\partial u}\Bigl[L(x,u)+V_{x}(x)\,f(x,u)\Bigr]=0
\]
Assumem que el cost instantani \(L\) i la dinàmica \(f\) són clàssiques (diferenciables) respecte al control \(u\).
---
#### 1. Escriu l’expressió conjunta
\[
\mathcal{H}(x,u,V_{x})\;=\;L(... |
daring_anteater_en-18909-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2026-10-06", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need provide code integration example. Provide minimal Flask app using functions from earlier. | 以下は、先ほど提示した `api_logger.py`（データベース操作・レートリミットロジック）を **Flask アプリに組み込む例**です。
ポイントは
1. **リクエストごとに `dummy_api` を呼び出す** ― これで自動的にログとレートリミット判定が走ります。
2. エラーハンドリングでレートリミット超過時は HTTP 429 を返す。
3. 必要ならエンドポイント名やクライアント ID の取得方法をカスタマイズできます（ここでは URL パラメータとヘッダーから取得）。
---
## 1️⃣ `api_logger.py`（そのまま使用）
```python
# api_logger.py... | We need provide code integration example. concise. | Below is a minimal Flask application that re‑uses the `api_logger.py` module you already have.
All requests go through a single decorator (`@track_and_throttle`) which:
1. extracts a client identifier (you can adapt this logic – here we use the remote IP);
2. checks the client’s rate‑limit via `check_allowed`;
3... |
daring_anteater_en-63452-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2027-11-09", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need show equivalence and compute limit using L'Hôpital and via Taylor. Provide concise answer. | **Why L’Hôpital’s rule is equivalent to a Taylor‐type expansion**
Let
\[
f(x)=\ln (1+x)-x ,\qquad g(x)=x^{2},
\]
and assume \(f(0)=g(0)=0\).
If we expand each function about \(x=0\),
\[
\begin{aligned}
\ln(1+x)&=x-\frac{x^{2}}{2}+ \frac{x^{3}}{3}-\cdots,\\[2mm]
g(x)&=x^{2}.
\end{aligned}
\]
Subtracting the lin... | We need show equivalence and compute limit via L'Hôpital and via Taylor. Provide concise answer. | **Why L’Hôpital’s rule is equivalent to using a Taylor expansion**
For a quotient
\[
\frac{f(x)}{g(x)},
\qquad f(0)=g(0)=0,\; g'(0)\neq0,
\]
the first‑order Taylor expansions about \(x=0\) are
\[
f(x)=f'(0)x+\tfrac12 f''(0)x^{2}+O(x^{3}),\qquad
g(x)=g'(0)x+\tfrac12 g''(0)x^{2}+O(x^{3}).
\]
Hence
\[
\frac... |
daring_anteater_en-90347-turn1--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2029-07-09", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | User wants discussion of decision-making levels effect on performance with resources, workforce, innovation. Provide structured answer. | **Strategic‑level Decision‑Making**
| Aspect | How it Shapes Performance | Key Mechanisms |
|--------|--------------------------|----------------|
| **Resource Allocation** | Sets the overall portfolio (capital, R&D budget, technology investments) that determines long‑term growth potential and competitive positioning.... | User wants discussion of decision-making levels effect on performance with resources, workforce, innovation. Provide concise but thorough explanation. | ### Decision‑Making Levels & Their Impact on Organizational Performance
| Level | Core Focus | Typical Decisions | Influence on Key Drivers |
|-------|------------|-------------------|--------------------------|
| **Strategic** | Long‑term direction, competitive positioning | • Vision/mission definition <br>• Market... |
daring_anteater_en-75762-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2026-07-09", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need give solution formulas. Provide general solution using diagonalization. Mention zero eigenvalue leads line of equilibria, not isolated node. Provide explicit x(t), y(t). | **Solution of the linear system**
\[
\begin{cases}
x'=-x+y \\[2pt]
y'= x-y
\end{cases},
\qquad
\mathbf u'(t)=A\,\mathbf u(t),\;
A=\begin{pmatrix}-1 & 1\\ 1 & -1\end{pmatrix}.
\]
--------------------------------------------------------------------
### 1. Diagonalisation
Eigenvalues of \(A\):
\[
\det(A-\lambda I)=... | We need give solution formulas. Provide general solution using eigen decomposition. Show dependence on constants. Explain effect of zero eigenvalue -> line of equilibria, not isolated node. Provide answer concise. | **Solution of the linear system**
\[
\begin{cases}
x'=-x+y \\[2pt]
y'=x-y
\end{cases},\qquad
\mathbf z'(t)=A\mathbf z(t),\;
A=\begin{pmatrix}-1&1\\[2pt]1&-1\end{pmatrix},
\qquad \mathbf z=(x,y)^T .
\]
Because \(A\) is constant we can diagonalise it.
------------------------------------------------------------------... |
daring_anteater_en-63166-turn2--1 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2029-07-24", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | Provide sample spreadsheet rows. | **Sample Spreadsheet Layout**
*(Columns are separated by tabs so you can copy‑paste directly into Excel, Google Sheets, or LibreOffice Calc)*
```
Saga Passage | Chapter / Verse | Translation (optional) | Claim About Raid | Archaeological Artifact ID | Site & Country | Publication (author, year) | Page / Fig. # | C... | Provide sample spreadsheet rows. | **Sample Spreadsheet Layout (CSV format)**
```
Saga Passage,Excerpt (Old Norse),English Translation,Archaeological Claim,Artifact ID / Site,Repository,Publication (Year),Confidence Level,Notes,Reference (Chicago footnote)
"Heimskringla – Harald Hardrada",«...þá ferðu um í Hestfjörð ok tóku þar býr…»,"...then they we... |
daring_anteater_en-13367-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2029-03-19", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need to answer question about neutrons in carbon isotopes and effect on stability. Provide concise explanation. | The number of neutrons in carbon isotopes is set when the nucleus forms during nuclear reactions (e.g., neutron capture, fission) or through natural processes such as cosmic‑ray spallation. Because all carbon atoms have six protons, changing only the neutron count creates distinct isotopes:
| Isotope | Protons | Neutr... | Need to answer question about neutrons determination and effect on stability. concise but thorough. | The number of neutrons in carbon isotopes is set when a nucleus forms, based on which neutron‑capture or decay event occurred during its creation (e.g., fusion in stars, cosmic‑ray spallation, nuclear reactions).
**How it’s determined**
* **Natural abundance:** Carbon‑12 (⁶C) results from stellar nucleosynthesis whe... |
daring_anteater_en-14734-turn2--0 | [{"role": "system", "name": null, "content": [{"model_identity": "You are LLM-jp-4, a large language model trained by LLM-jp.", "reasoning_effort": "Low", "conversation_start_date": "2029-06-09", "knowledge_cutoff": "2025-12", "channel_config": {"valid_channels": ["analysis", "commentary", "final"], "channel_required":... | We need give concrete activities for each guidance item listed. List per question. Provide concise actionable steps. Ensure inclusive language. | Below are ready‑to‑use activities that correspond to the “Guidance / Area for Improvement” column in the answer key.
Each activity includes a short description, step‑by‑step instructions, estimated time required, and tips for adapting it to different abilities, languages, or cultural contexts.
| # | Target skill (fr... | We need give concrete examples for each guidance action listed. Provide list per question/action. concise. | Below are ready‑to‑use activities that map directly onto the “Guidance / Area for Improvement” suggestions in the answer key.
Pick any activity that matches the domain(s) you wish to strengthen, schedule it consistently (daily, weekly, or as noted), and note your experience in a short journal entry (“what worked, wha... |
End of preview. Expand in Data Studio
llm-jp-4.1-32b-a3b-thinking-dpo-data
Overview
This dataset is a Direct Preference Optimization (DPO) dataset used to train llm-jp-4.1-32b-a3b-thinking.
It is constructed by pairing multiple candidate responses for a given prompt and selecting preferred (chosen) and non-preferred (rejected) responses. The splits reasoning_low, reasoning_medium, and reasoning_high correspond to different reasoning effort settings used during response generation.
The fields chosen_analysis, chosen_final, rejected_analysis, and rejected_final are generated using a model trained via supervised fine-tuning (SFT). These responses are then evaluated by gpt-oss-120b, which determines the preferred (chosen) and non-preferred (rejected) responses.
To support the continued development of LLM-jp, we would greatly appreciate it if you could share how you utilize LLM-jp outcomes via the survey form.
Data Sources and Licenses
The prompts used to construct this DPO dataset are derived from the following data sources. Since each source dataset has its own license, please ensure compliance with the respective licenses when using this dataset. Some subsets cannot be redistributed due to licensing restrictions and are therefore not included.
| Dataset | License | Notes |
|---|---|---|
| ac_self_inst | answer-carefully-dataset-tou | Not included due to redistribution restrictions |
| daring_anteater | CC BY 4.0 | - |
| flan | ODC-BY | - |
| ichikara | - | Requires paid license, not included |
| jaster_v1.4.1 | CC BY-SA 4.0, CC BY-SA 3.0, CC BY 4.0, BSD-3, Apache-2.0, MIT | - |
| llmjp_extraction_wiki_ja_v0.x | Apache-2.0 | - |
| llmjp_magpie_sft_v1.0 | Apache-2.0 | - |
| llmc_math_dataset | - | To be released |
| logical_math_coding_wizard8x22b | Apache-2.0 | - |
| multiturn_calm3 | Apache-2.0, CC BY-SA 3.0, CC0, CC BY 4.0 | - |
| nemotron_post_v2_stem | CC BY 4.0 | - |
| nemotron_post_v3_chat | CC BY 4.0 | - |
| nemotron_post_v3_if | CC BY 4.0 | - |
| nemotron_post_v3_math | CC BY-SA 4.0 | - |
| random_to_fixed_multiturn_calm3 | Apache-2.0 | - |
| synthetic_if_ja | - | To be released |
| synthetic_jp_en_coding | Apache-2.0 | - |
| nemotron3_sft_chat_v3 | CC BY 4.0, ODC-BY | - |
| nemotron3_sft_if_v3 | CC BY 4.0, ODC-BY | - |
| nemotron3_sft_science_vendor | CC BY-SA 4.0 | - |
| table_gpt_train_large | MIT | - |
| nemotron3_sft_multilingual_v2_math_ko_stackoverflow | CC BY-SA 4.0 | - |
| nemotron3_sft_multilingual_v2_math_ja_stackoverflow | CC BY-SA 4.0 | - |
Data Format
Each sample has the following format:
{
"ID": "...",
"messages": [
{"role": "system", "name": null, "content": ["..."]},
{"role": "user", "name": null, "content": ["..."]},
],
"chosen_analysis": "...",
"chosen_final": "...",
"rejected_analysis": "...",
"rejected_final": "..."
}
chosen_analysisrepresents the reasoning process (chain-of-thought) of the preferred responsechosen_finalrepresents the final response of the preferred responserejected_analysisrepresents the reasoning process of the non-preferred responserejected_finalrepresents the final response of the non-preferred response
The messages field contains the input conversation (prompt context).
This structure allows training preference models that compare two candidate responses conditioned on the same prompt.
Notes
- This dataset contains responses generated by a large language model.
- The preference labels (chosen vs rejected) are constructed automatically and may contain noise.
- Generated content may include errors, inaccuracies, biases, or harmful outputs.
Send Questions to
llm-jp(at)nii.ac.jp
Model Card Authors
The names are listed in alphabetical order.
Hirokazu Kiyomaru and Takashi Kodama.
- Downloads last month
- 1,828
