Rebiha

Home › Datasets › Science & Research

Science & Research Fine-Tuning Dataset

Scientific research, methods, analysis, peer review

Fine-tune a model on this dataset

$19 for the dataset. Training is priced separately by model, starting at $47.

What is in this dataset

936 instruction-response pairs covering scientific research, methods, analysis, peer review. Each pair is a realistic question with a full written answer, in the JSONL format used for supervised fine-tuning. No synthetic filler, no duplicated rows.

Example pairs

Real rows from the dataset, shown in full.

Design a controlled experiment to test whether a new fertilizer increases plant growth compared to a standard fertilizer. Specify the control group, independent/dependent variables, and how you'd minimize bias.

To test the new fertilizer, set up a randomized controlled experiment. Use two groups: an experimental group receiving the new fertilizer and a control group receiving the standard fertilizer (or no fertilizer as a baseline). The independent variable is the type of fertilizer; the dependent variable is plant growth, measured as height, biomass, or leaf count after a fixed period. Minimize bias by using identical plants (same species, age, and initial size), planting in identical soil and pots, providing equal light, water, and temperature, and randomizing assignment to groups. Use blinding: the person measuring growth should not know which group a plant belongs to. Replicate with at least…

Explain the difference between a one-tailed and two-tailed t-test, and give a concrete example of when to use each.

A t-test compares means between two groups. A one-tailed t-test tests for a difference in a specific direction (e.g., Group A > Group B or Group A < Group B). Use it when you have a strong prior hypothesis that the effect will only go one way. Example: testing if a new drug lowers blood pressure more than a placebo; you only care if it's lower, not higher. A two-tailed t-test tests for any difference (Group A ≠ Group B) without direction. Use it when you cannot rule out the possibility of an effect in either direction. Example: comparing test scores from two teaching methods; you don't know which might be better, so you test for any difference. The two-tailed test is more…

I ran an experiment and got a p-value of 0.03. Does this mean there's only a 3% chance the null hypothesis is true? Explain how to correctly interpret this p-value.

No, that is a common misinterpretation. A p-value of 0.03 means that if the null hypothesis were true, there is a 3% probability of observing data as extreme as yours (or more extreme) purely due to random chance. It does not say anything about the probability that the null hypothesis itself is true; that would require a Bayesian approach. Correct interpretation: assuming the null hypothesis is true, the observed result is unlikely (less than 5% threshold), so you have evidence to reject the null hypothesis. However, p-values do not measure effect size, practical significance, or the probability of replication. Always report the effect size and confidence interval alongside the p-value to…

Who this is for

Research assistants covering methods, statistics, and peer review.

Which models this works with

The dataset is plain JSONL, so it works with any instruction-tuned open model. On Rebiha you can train it directly on:

Training uses LoRA, or QLoRA on a 4-bit base for larger models. Your base model's weights are never modified — training produces a separate adapter.

How it works

  1. Pick this dataset and a base model.
  2. Rebiha runs the job on a managed GPU. You set up nothing.
  3. Download your model: a ready-to-run GGUF, plus developer assets (adapter, tokenizer, configs).

Fine-tuning or RAG?

Use RAG when the model needs facts that change — your prices, your documents, today's inventory. Use fine-tuning when you want consistent style and structure without prompting for it every time. This dataset teaches the shape of a good answer in this domain, not facts to look up.

Related datasets

Start fine-tuning