Poro 2 Long
Collection
5 items • Updated
id int64 27 2.23M | messages listlengths 2 2 | translated_messages listlengths 2 2 | translated_traces_success bool 2
classes | translator stringclasses 1
value | original_source stringclasses 1
value |
|---|---|---|---|---|---|
466,957 | [
{
"role": "user",
"content": "Solve the following math problem. Make sure to put the answer (and only answer) inside \\boxed{}.\n\nEvaluate the limit:\n\n\\[ \\lim_{n \\to \\infty} \\left( \\sum_{k = 0}^{n} \\frac{1}{\\binom{n}{k}} \\right) \\]\n\nwhere \\(\\binom{n}{k} = \\frac{n!}{k!(n-k)!}\\)."
},
{
... | [
{
"role": "user",
"content": "Ratkaise seuraava matematiikan tehtävä. Varmista, että vastaus (ja vain vastaus) sijoitetaan \\boxed{}-sulkeiden sisään.\n\nArvioi raja-arvo:\n\n\\[ \\lim_{n \\to \\infty} \\left( \\sum_{k = 0}^{n} \\frac{1}{\\binom{n}{k}} \\right) \\]\n\nmissä \\(\\binom{n}{k} = \\frac{n!}{k!(... | true | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
104,903 | [
{
"role": "user",
"content": "Solve the following math problem. Make sure to put the answer (and only answer) inside \\boxed{}.\n\nGive the exact formula for the minimum depth of a $d$-ary tree with $n$ vertices."
},
{
"role": "assistant",
"content": "<think>\nOkay, so I need to find the exact f... | [
{
"role": "user",
"content": "Ratkaise seuraava matemaattinen ongelma. Varmista, että vastaus (ja vain vastaus) sijoitetaan \\boxed{} sisään.\n\nAnna tarkka kaava $d$-aarisen puun minimisyvyydelle, jossa on $n$ kärkeä."
},
{
"role": "assistant",
"content": "<think>\nOkei, joten minun täytyy löyt... | true | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
1,153,557 | [
{
"role": "user",
"content": "Solve the following math problem. Make sure to put the answer (and only answer) inside \\boxed{}.\n\nLet \\( f \\in \\mathbb{C}[X] \\) be a polynomial of degree \\( n \\geq 0 \\). What happens when \\( r < 1 \\) regarding the representation of \\( f \\) as a sum of polynomials ... | [
{
"role": "user",
"content": "Ratkaise seuraava matematiikan ongelma. Varmista, että vastaus (ja vain vastaus) on laatikossa \\boxed{}.\n\nOlkoon \\( f \\in \\mathbb{C}[X] \\) polynomi, jonka aste on \\( n \\geq 0 \\). Mitä tapahtuu, kun \\( r < 1 \\) polynomin \\( f \\) esitykselle polynomien summana, joid... | false | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
1,027,151 | [
{
"role": "user",
"content": "Solve the following math problem. Make sure to put the answer (and only answer) inside \\boxed{}.\n\nSolve in $\\mathbb{R}$ the equation:\n\\[\n\\log_{278}(\\sqrt{x} + \\sqrt[4]{x} + \\sqrt[8]{x} + \\sqrt[16]{x}) = \\log_2(\\sqrt[16]{x}).\n\\]"
},
{
"role": "assistant",... | [
{
"role": "user",
"content": "Ratkaise seuraava matemaattinen ongelma. Varmista, että vastaus (ja vain vastaus) sijoitetaan \\boxed{} sisään.\n\nRatkaise joukossa $\\mathbb{R}$ yhtälö:\n\\[\n\\log_{278}(\\sqrt{x} + \\sqrt[4]{x} + \\sqrt[8]{x} + \\sqrt[16]{x}) = \\log_2(\\sqrt[16]{x}).\n\\]"
},
{
"ro... | true | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
936,214 | [{"role":"user","content":"Solve the following math problem. Make sure to put the answer (and only a(...TRUNCATED) | [{"role":"user","content":"Ratkaise seuraava matematiikan tehtävä. Varmista, että vastaus (ja vai(...TRUNCATED) | true | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
585,265 | [{"role":"user","content":"Solve the following math problem. Make sure to put the answer (and only a(...TRUNCATED) | [{"role":"user","content":"Ratkaise seuraava matemaattinen tehtävä. Varmista, että vastaus (ja va(...TRUNCATED) | true | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
429,896 | [{"role":"user","content":"Solve the following math problem. Make sure to put the answer (and only a(...TRUNCATED) | [{"role":"user","content":"Ratkaise seuraava matemaattinen ongelma. Varmista, että vastaus (ja vain(...TRUNCATED) | false | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
364,648 | [{"role":"user","content":"Solve the following math problem. Make sure to put the answer (and only a(...TRUNCATED) | [{"role":"user","content":"Ratkaise seuraava matemaattinen ongelma. Varmista, että vastaus (ja vain(...TRUNCATED) | false | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
1,769,669 | [{"role":"user","content":"Solve the following math problem. Make sure to put the answer (and only a(...TRUNCATED) | [{"role":"user","content":"Ratkaise seuraava matemaattinen ongelma. Varmista, että vastaus (ja vain(...TRUNCATED) | false | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
133,307 | [{"role":"user","content":"Solve the following math problem. Make sure to put the answer (and only a(...TRUNCATED) | [{"role":"user","content":"Ratkaise seuraava matemaattinen ongelma. Varmista, että vastaus (ja vain(...TRUNCATED) | true | DeepSeek-V3 | Llama-Nemotron-Post-Training-Dataset_SFT_math |
End of preview. Expand in Data Studio
Llama-Nemotron-Post-Training-Dataset-SFT-math-FI
This dataset is a Finnish machine-translated version of the SFT/math split from the original nvidia/Llama-Nemotron-Post-Training-Dataset.
The data was created by translating the original English math SFT subset into Finnish using the DeepSeek-V3 model.
Translation Process
The user prompt and the thinking traces were translated separately in two LLM requests. For traces, the <think> and </think> tokens were preserved in their original positions.
The following prompt was used for both LLM translation requests:
You are a professional translator specializing in mathematics and scientific texts. Your task is to translate the following content faithfully and accurately into {language}.
Guidelines:
1. Preserve all LaTeX code, equations, symbols, and formatting exactly as written.
2. Translate only the surrounding natural language, not the math expressions inside \( ... \), \[ ... \], or $$ ... $$.
3. Maintain the precise meaning, tone, and logical structure of the original text.
4. Use the standard mathematical terminology of the target language.
5. Do not simplify, interpret, or solve the math; your goal is linguistic translation only.
6. Keep variable names, constants, and notation unchanged.
7. If an English math term has multiple valid equivalents in the target language, choose the most widely accepted in academic usage.
8. Do not explain your translation; output only the translated text unless asked otherwise.
9. Preserve the `<think>` and `</think>` tags in the resulting text in the same positions as in the original input.
Text to translate:
{text}
Post-processing and Filtering
After translation, heuristic checks were applied to filter out incomplete or malformed translation outputs.
A translated trace was kept only if:
- The translation starts with
<think>. - The translation contains exactly one
<think>tag and exactly one</think>tag. - There is non-whitespace content after the closing
</think>tag, since the final answer is expected to appear after the thinking trace. - The token count difference between the original English trace and the translated Finnish trace is no more than 5,000 tokens.
A translated trace was considered successful if it met all of the above criteria. Rows that failed these checks were not removed from this dataset — they are retained and flagged via the translated_traces_success column so that users can filter them out if only clean translations are needed.
Dataset Statistics
- Total samples: 2,202,426
- Successfully translated (
translated_traces_success = True): 2,101,293 (95.41%) - Failed post-processing checks (
translated_traces_success = False): 101,133 (4.59%)
Breakdown of failure reasons, as a percentage of the total dataset (not just of the failed subset):
| Failure reason | Count | % of total dataset |
|---|---|---|
Missing closing </think> tag (thinking trace truncated) |
54,118 | 2.457% |
Both tags present, but no content after </think> (final answer truncated) |
46,136 | 2.095% |
| Token count difference between original and translation exceeds 5,000 | 658 | 0.030% |
Malformed tag structure (e.g. extra </think> tags) |
143 | 0.006% |
Does not start with <think> |
78 | 0.004% |
Nearly all failures (99.3%) stem from the DeepSeek-V3 generation being cut off before completing the thinking trace or the final answer, rather than from translation quality issues.
Content length also increases moderately from English to Finnish: the translated assistant responses are on average about 9% longer in characters than the original English responses (mean 18,175 vs. 17,254 characters; based on the assistant turn of each sample).
License
This dataset is released under CC-BY-4.0, consistent with the license of its source, the SFT/math split of nvidia/Llama-Nemotron-Post-Training-Dataset, which is predominantly CC-BY-4.0 (the source dataset notes license is tracked per-sample, with only small non-math subsets under other terms).
This dataset was translated using DeepSeek-V3, which is governed by the DeepSeek License Agreement. DeepSeek claims no rights over model outputs, but use of the output remains subject to the Use Restrictions in Attachment A of that license (e.g. no military use, no generation of harmful or discriminatory content).
- Downloads last month
- 5