Sft

Match each training stage to its objective and data signal

PremiumWhy does SFT struggle…

Identify the objective mistake in this pretraining to SFT claim

PremiumWhy does completion only…

Which axes of diversity matter most in Tulu style instruction tuning?

PremiumSpot the bug: packed…

PremiumWhat is the root…

Match each instruction dataset to its on disk row shape.

Give one training example for SFT'ing a single weather tool call.

How much general data should a replay buffer for domain SFT use, and from which general dataset?

How should domain and general data balance in a code assistant SFT mix?

What is teacher forcing during SFT?

Should the system message contribute to the SFT loss?

Fill in the SFT JSONL: the three required fields of a typical chat example.

What is supervised fine-tuning (SFT)?

What are the three standard roles in a chat training example?

How do pretraining and fine-tuning differ in data and in the objective?

What is the sensible num_train_epochs for a typical instruction SFT run on a curated dataset?

Which loss function does supervised fine-tuning minimize?

How does a base model differ from an instruction tuned model?

What are the three fields in an Alpaca format example?

What travels from teacher to student in DeepSeek style reasoning distillation, and what gets cut from the loss?

Compare SFT and continued pretraining when adapting a base model to medical text

Spot the error in this description of SFT loss masking

Explain SFT's loss and the role of prompt token masking