Home › Datasets › General Instruction Following
General Assistant PackHow to Fine-Tune an LLM for Better Instruction Following
Following clear and ambiguous instructions accurately — formatting, constraints, multi-step requests, and precise task completion.
- 978 instruction-response pairs
- Format JSONL
- Pack General Assistant
- Dataset $19 one-time
$19 for the dataset. Training is priced separately by model, starting at $47.
What can this kind of fine-tuning teach an LLM?
A general-purpose model may already know some of this subject. Fine-tuning serves a different purpose: it repeatedly demonstrates the kinds of requests, response patterns, terminology, decisions and workflows you want the model to handle more consistently.
For General Instruction Following, the dataset focuses on following clear and ambiguous instructions accurately — formatting, constraints, multi-step requests, and precise task completion. Fine-tuning shapes recurring model behavior; it does not turn the model into a live database or guarantee correctness on every new case.
What training data should cover
Useful supervised fine-tuning data should resemble the work you expect the model to perform. This Rebiha dataset's domain definition is organized around 20 coverage areas:
- following a strict word or character limit exactly; producing output in a specific format (list, table, JSON); following a multi-step instruction in the given order; adhering to a specified tone or persona; handling a instruction with multiple constraints at once
- correctly interpreting an ambiguous instruction with a reasonable default; following a negative constraint (avoid X, don't mention Y); producing exactly N items when asked for a specific count; following formatting instructions for headers and structure; translating a vague request into a well-scoped response
- handling conflicting instructions sensibly; following a step-by-step process instruction precisely; producing a response within a specific reading level; following an instruction to exclude certain content; adapting output length to match instruction (brief vs detailed)
- correctly following an instruction with nested sub-requirements; producing a response in a specified language or register; following an instruction that specifies audience; handling an instruction that changes mid-conversation; confirming understanding before proceeding on a complex ask
How Rebiha prepared the training material
This page contains 978 instruction-response pairs from the existing Rebiha dataset. Its domain definition was designed around the coverage areas above so the training material spans multiple realistic tasks rather than one narrow prompt pattern.
Rebiha's broader dataset workflow uses structured quality review to help identify problems such as contradictions, unsupported claims, unsupported assumptions, calculation errors and other weak training examples. You can inspect real examples on this page before deciding whether the dataset fits your model and use case.
What is in this dataset
978 instruction-response pairs covering following clear and ambiguous instructions accurately — formatting, constraints, multi-step requests, and precise task completion. Each pair is a realistic instruction with a full written response, 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.
Write a one-sentence product description for a reusable water bottle that is exactly 25 words long. Count the words carefully—do not exceed or fall short.
This insulated steel bottle keeps drinks cold for twenty-four hours, fits most cup holders, and is dishwasher safe. Its leak-proof lid makes every sip effortless.
List the steps to change a flat tire on a car, but present them as a numbered list with exactly 6 items. Each item must be a single sentence starting with a verb. Do not add extra commentary.
1. Park the vehicle on level ground and engage the parking brake. 2. Loosen the lug nuts slightly with the wrench while the tire is still on the ground. 3. Place the jack under the vehicle's designated lift point and raise it until the tire clears the ground. 4. Remove the loosened lug nuts and pull the flat tire straight off the wheel hub. 5. Mount the spare tire onto the hub and thread the lug nuts back on by hand until snug. 6. Lower the vehicle, then tighten the lug nuts in a star pattern using the wrench.
You are a customer service bot for an airline. A passenger asks: 'My flight is delayed 3 hours, can I get a refund?' Respond in JSON format with three keys: "refund_eligible" (boolean), "policy_explanation" (string, max 40 words), and "next_step" (string). Follow the airline policy: refunds are only allowed if the delay exceeds 4 hours; otherwise offer meal voucher.
{"refund_eligible": false, "policy_explanation": "Our policy grants refunds only for delays over four hours. Since your delay is three hours, a refund is not available.", "next_step": "Please visit our service desk to receive a meal voucher for the delay."}
Who this is for
Builders who want their fine-tuned model to reliably perform this specific capability, as part of a broader assistant or agent.
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:
- Qwen — Qwen3.8, Qwen3.6, Qwen3.5, Qwen3, Qwen2.5, Qwen2.5-Coder
- Granite — Granite 4.1 (3B, 8B, 30B)
- OLMo — OLMo 3, OLMo 3.1
- GLM — GLM-4
- Gemma, Phi, DeepSeek, Mistral, and Llama families
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
- Pick this dataset and a base model.
- Rebiha runs the job on a managed GPU. You set up nothing.
- 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 behavior and structure without prompting for it every time. This dataset teaches the shape of this capability, not facts to look up.
How managed fine-tuning works
- Inspect the dataset examples and choose a compatible base model.
- Rebiha runs the training job on managed GPU infrastructure using LoRA or QLoRA as appropriate.
- Download portable model assets for your own workflow, including the available ready-to-run and developer outputs for the selected training path.
Fine-tuning or RAG?
Use fine-tuning when the objective is recurring behavior: response structure, domain patterns, terminology, workflow, or how the model approaches a class of requests. Use RAG when answers depend on changing or proprietary facts that should be retrieved at request time. The two approaches can also work together.
Know what goes in. Own what comes out.
You can inspect real examples before training, choose the model you want to adapt, and keep portable model assets when training is complete.