Methods › General › Distillation › SFT

Shrink and Fine-Tune

SFT

415 papers tagged archive 2025-07-28

Introduced by Sam Shleifer et al. in Pre-trained Summarization Distillation

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

Shrink and Fine-Tune, or SFT, is a type of distillation that avoids explicit distillation by copying parameters to a student student model and then fine-tuning. Specifically it extracts a student model from the maximally spaced layers of a fine-tuned teacher. Each layer l ∈L′ is copied fully from L. For example, when creating a BART student with 3 decoder layers from the 12 encoder layer 12 decoder layer teacher, we copy the teacher’s full Encᴸ and decoder layers 0, 6, and 11 to the student. When deciding which layers to copy, we break ties arbitrarily; copying layers 0, 5, and 11 might work just as well. When copy only 1 decoder layer, we copy layer 0. This was found this to work better than copying layer 11. The impact of initialization on performance is measured experimentally in Section 6.1. After initialization, the student model continues to fine-tune on the summarization dataset, with the objective of minimizing ℒ_(Data).

PaperSource

In Syntology the paper matched by title, the paper Syntology's graph keys as openreview_m4eD6HDGGX, held, no code (no_code; paper_only). The lookup matched a paper by its title, the paper Syntology's graph keys as openreview_m4eD6HDGGX, not a method entry. The archive credits no paper with an arXiv id for this method, so whether this answer (about the paper Syntology's graph keys as openreview_m4eD6HDGGX) is about this method is not checked; it does not say whether this method has code.

From the lookup's note on this name: The same name also belongs to “Self-Filtering (SFT)”, a method in the paper “Self-Filtering: A Noise-Aware Sample Selection for Label Noise with Confidence Penalization” (arXiv:2208.11351, 64 citations), which the lookup finds cited far more than the paper this answer is about (citations not recorded), so this answer may be about a different sense of the name. For arXiv:2208.11351: its page on this site · ask the lookup about it.

Syntology's have check for this name (free, it answers coverage only), read 2026-10-05; ask it again live. A level is the served verification level of that one answer; nothing here is summed across methods.

Papers archive 2025-07-28

archive papers tagged: 415 · with a code link: 204 · where Syntology ran a sample: 103 (82 with a run with no instrument failure, 21 where every run was a failure of Syntology's instrument) Syntology

Show: all tagged papersonly where code ran (103 of 415 tagged: 82 with a run with no instrument failure, 21 where every run was a failure of Syntology's instrument)

Page 1 of 5: papers 1 to 100 of 415, newest first by the archive's date (ties by slug), in archive order.

Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code, as “N ran (of which C constructed an object rather than computing a result; K with no instrument failure: H honoured, V violated, P with no contract checked; I where Syntology's instrument failed) · U unverified”; the instrument figure counts failures of Syntology's instrument, not of the code. It is per sample and not a correctness claim. When the archive marks a repository official for the paper, the line starts with that repository's state (the archive's flag, not a verdict on who wrote the code; “community repositories only” when every sample that ran came from a community repository, “official: no sample here; runs from other or unrecorded repositories” when some came from a repository the paper names or has in its text, or from none recorded); hover it for the repositories the samples that ran came from.

Tasks archive 2025-07-28

20 shown of 206 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Reinforcement Learning (RL)73
Language Modelling50
Language Modeling41
Large Language Model35
Instruction Following34
Question Answering30
Math29
Diversity23
Mathematical Reasoning21
Reinforcement Learning20
reinforcement-learning20
In-Context Learning17
Code Generation13
GSM8K13
Retrieval11
RAG10
Text Generation10
Visual Reasoning10
Retrieval-augmented Generation9
Machine Translation8

Usage over time archive 2025-07-28

Papers per year tagged with SFT: 2020 to 2025, peak 187 187 0 2020: 1 paper 2020 2021: 0 papers 2021 2022: 3 papers 2022 2023: 41 papers 2023 2024: 187 papers 2024 2025: 183 papers 2025
Papers per year the archive tags with this method, by the paper's archive date (415 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

DistillationKnowledge Distillation

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections