In a DeepSeek-style reasoning distillation recipe, describe what the teacher actually produces, what shape the student's training examples take, which tokens are inside the loss versus masked out, and why the distillation collapses if the chain-of-thought is dropped from the target.
Free · 2 AI evals / day
**The teacher does more than answer.** A strong reasoning model is prompted on hard problems and produces a full step by step reasoning trace, followed by a final answer. Both parts go into the training data.
Each example for the student is a triple of (problem, trace, answer). When you format it as a single sequence for SFT, the problem is the prompt and the trace plus answer together are the response.
The loss is standard completion-only cross-entropy. Prompt tokens are masked with -100 so they do not contribute to the gradient. Trace and answer tokens are kept inside the loss. The student is graded on reproducing the teacher's full working, not just the bottom line.
If you drop the trace and supervise on (problem, answer) pairs, the student learns to map problems to answers on the training distribution and not much else. The procedure that produces those answers is missing from the signal, so the student cannot generalise to a new problem. The trace is what carries the reasoning ability across.
**The teacher produces traces, not just answers.** A strong reasoning model, prompted on hard problems, emits a full chain-of-thought followed by a final answer. Each training example becomes a (problem, trace, answer) triple, formatted as one sequence in the chat template: the problem sits in the prompt span, the trace plus answer sits in the response span.
The loss is completion-only SFT cross-entropy. Prompt tokens are masked with the ignore_index of -100. Trace tokens and final answer tokens are kept in the loss. The student is graded on producing the same reasoning trajectory the teacher used, token by token.
Quality control usually filters by correctness. Only traces whose final answer matches the gold answer are kept; the rest are dropped or rejection-sampled. Some recipes also reweight examples or run a brief preference-tuning pass on top, but the core SFT step is the load-bearing one.
Dropping the trace collapses the recipe. With (problem, answer) supervision alone, the student learns a shallow mapping from problems to answers on the training distribution. The procedure that produced those answers is invisible to it. A held-out problem looks new, and there is no learned reasoning template to apply. You end up with a model that memorises a fixed set of answers rather than one that has absorbed the teacher's reasoning ability.
Quick reference
TL;DR
Teacher emits trace plus answer. Each example is a (problem, trace, answer) triple; SFT loss covers trace and answer with the prompt masked. Drop the trace and you teach answer-copying, not reasoning.
Imagine learning math from a tutor in two different styles. In style one, the tutor solves problems on a whiteboard step by step, and you copy the whole working including every line of arithmetic. In style two, the tutor only writes the final answer at the bottom and erases the rest. After a week of style one, you can solve new problems because you internalised the procedure. After a week of style two, you can recite the answers to the problems you saw, but a new problem stumps you because you never saw how the tutor got there. DeepSeek-style distillation works because the student copies the whole whiteboard, including the messy middle. Skip the middle and the student learns to mimic the bottom line of a fixed list of problems.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
Tasks with no meaningful intermediate reasoning, like single-turn classification or short factual lookup There is no trace worth distilling; (problem, answer) SFT is sufficient and trace generation only adds cost.
The available teacher cannot produce reliable traces and traces fail correctness filtering at high rates A noisy or wrong trace corpus teaches the student bad reasoning. You inherit the teacher's failure modes amplified by the SFT objective.
Tight latency budget that cannot afford long generations at inference time A trace-distilled student will tend to produce long chain-of-thought outputs at inference, which can be slow and expensive without careful prompting or post-training to suppress traces.
Quick scan · 2–4 min
Everything important, quickly.
5 min: teacher emits trace plus answer + triple as data shape + completion-only masking with trace in loss + correctness filtering + why trace in loss carries reasoning + comparison to answer-only SFT.
Real-world examples
Real products, models, and research that use this idea.
1DeepSeek's R1 distillation release publicly demonstrated this recipe by distilling reasoning behavior from R1 into smaller dense models, with the full chain-of-thought kept inside the supervised target.
2Open-source replications such as OpenThoughts and DeepSeek V4 Reasoning distill curated trace plus answer datasets from strong teachers like Claude Opus 4.7 or GPT-5.5 into 7B to 70B students.
3Hugging Face's TRL SFTTrainer with completion-only masking is the standard implementation surface; the trace plus answer is wrapped in the assistant turn so the masking is automatic.
4Microsoft's Orca and Phi families pioneered the trace-distillation pattern earlier, showing measurable reasoning lift from including step by step traces in the supervised target.
Interviewer follow-ups
What an interviewer would ask next. Try answering before peeking at the approach.
QHow does trace distillation compare to logit distillation on the same teacher?
Sign in or go Pro to see the approach for this follow-up.
QWhy does correctness filtering matter even when the teacher is strong?
Sign in or go Pro to see the approach for this follow-up.
QWhat changes if the student is then preference-tuned on top of the SFT step?
Sign in or go Pro to see the approach for this follow-up.
Don't say this
Red flags and common mistakes that signal junior thinking. Click to expand.
Most common mistake
Treating the trace as part of the prompt rather than part of the response. The trace must sit inside the loss; the student is being graded on reproducing it, not conditioning on it.