Skip to content
finetuning-method-selection ยท wshobson-agents ยท Flowy
Decide whether to fine-tune at all, and route to the right method (SFT, DPO/ORPO/KTO, GRPO/RLVR, continued pretraining) and base model. Use when starting any fine-tuning effort, when unsure whether RAG or prompting would suffice, or when choosing between preference-optimization
One skill from wshobson-agents .
$ npx -y skills add wshobson/agents --skill finetuning-method-selection --agent claude-codeInstalls just this skill. Get the whole plugin for auto-invocation.
โก How it fires
How this skill gets triggered: by you, by Claude, or both.
Fires itself Claude auto-loads it when your prompt matches the work. You can call it Invoke it directly when you want it. Slash command /finetuning-method-selection๐๏ธ Context preview
The summary Claude sees to decide when to auto-load this skill.
Decide whether to fine-tune at all, and route to the right method (SFT, DPO/ORPO/KTO, GRPO/RLVR, continued pretraining) and base model. Use when starting any fine-tuning effort, when unsure whether RAG or prompting would suffice, or when choosing between preference-optimization ๐ฆ Ships with wshobson-agents
</> SKILL.md
finetuning-method-selection.SKILL.md
---
name : finetuning-method-selection
description : Decide whether to fine-tune at all, and route to the right method (SFT, DPO/ORPO/KTO, GRPO/RLVR, continued pretraining) and base model. Use when starting any fine-tuning effort, when unsure whether RAG or prompting would suffice, or when choosing between preference-optimization and reinforcement methods.
---
# Fine-Tuning Method Selection
This is the router skill for the fine-tuning
lifecycle : it decides whether fine-tuning is the
right tool at all, and if so, which method and
which base-model size class. Every other skill
in this plugin assumes this routing already
happened โ start here before opening
`lora-qlora-recipes`, `preference-optimization`,
or `grpo-rlvr-training`.
## When to Use This Skill
- Starting any fine-tuning effort, before a
framework or base model has been chosen.
- Unsure whether RAG or prompt engineering would
solve the problem more cheaply than training.
- Choosing between preference optimization (DPO
family) and a reinforcement method (GRPO/RLVR)
for the same underlying task.
- Sizing a candidate model/method combination
before committing to a run.
## Quick Reference
Flowy AI Flows that just work. Hand-picked Claude Code plugins that fire the right skill as you prompt.
| Situation | Route |
|---|---|
| Facts change often (prices, docs, news) | RAG, not fine-tuning |
| Desired behavior still being figured out | Prompt engineering |
| Stable domain knowledge, โฅ500MB text | CPT then SFT โ see Off-Ramps First |
| Have input/output demonstrations | SFT โ see `lora-qlora-recipes` |
| Have preference pairs or thumbs-up/down | DPO/ORPO/KTO โ see `preference-optimization` |
| Have a verifiable pass/fail signal | GRPO+RLVR โ see `grpo-rlvr-training` |
| No eval harness yet | Stop โ see `eval-harness-first` |
## Off-Ramps First
Most requests that sound like "fine-tune this"
are served better and cheaper elsewhere. Check
these off-ramps before opening a training run:
- **Knowledge-bound and volatile** (the gap is
facts that change โ prices, docs, current
events): route to RAG, not fine-tuning. A
fine-tuned model bakes in a snapshot; volatile
facts go stale immediately.
- **Behavior-bound and shifting** (the desired
behavior is still being figured out, or
changes per request): route to prompt
engineering. Fine-tuning locks in a behavior;
don't lock in one that hasn't stabilized yet.
- **Stable, dense domain knowledge**: this is
where continued pretraining (CPT) enters, sized
by how much domain text exists:
| Domain text volume | Route |
|---|---|
| <10MB | RAG only |
| 10MBโ500MB | RAG + fine-tune |
| 500MBโ10GB | CPT, then SFT |
| >10GB | CPT required |
CPT learning rate โ **10% of the pretraining
LR**. CPT is guidance-only in this plugin โ
sizing and LR guidance live here, but this
plugin does not execute a CPT run.
## Method Router
Once the off-ramps are ruled out, this is the
full decision tree (verbatim from the research
this plugin is built on):
```
New FACTS? volatile โ RAG | stable+dense โ CPT (LR ~10% of pretrain) โ SFT
New BEHAVIOR? shifting โ prompt-engineering | stable:
demos โ SFT (LoRA/QLoRA, all-linear, ฮฑ=2r)
preference pairs โ DPO (SimPO if length-bias, ORPO if memory-bound)
unpaired ๐/๐ โ KTO
verifiable success โ RLVR + GRPO (DAPO/GSPO/Dr.GRPO per failure mode)
Deploy: FP8 (Hopper+) | NVFP4 (Blackwell scale) | AWQ (older) | GGUF+imatrix (edge)
BEFORE ANY OF THIS: the eval harness must exist first.
```
Read the tree top-down: answer "new facts or new
behavior," then follow the branch that matches
the data shape in hand (demos, preference pairs,
thumbs up/down, or verifiable success/failure).
The data shape picks the method โ not the other
way around.
### Worked Routing Examples
- *"Users want the assistant to follow our
support macros exactly."* Behavior is stable
and demonstrable from transcripts โ demos โ
**SFT**.
- *"We have pairs of good/bad responses from
reviewer thumbs-up/down, unpaired."* โ unpaired
signal โ **KTO**, not DPO (DPO needs paired
preferences).
- *"The model can already solve some of these
math problems and we can grade correctness
automatically."* โ verifiable success signal โ
**GRPO+RLVR**, and only after confirming the
model succeeds at least sometimes (see Key
Routing Facts below).
- *"We want the model to know this week's
pricing page."* โ volatile facts โ **RAG**, no
training run at all.
## Key Routing Facts
- **Loss-function choice is low-leverage.** A
240-H100-run study found method choice worth
~1 percentage point versus ~50 points for model
scale, and zero of 20 DPO variants beat vanilla
DPO. Don't spend a routing decision agonizing
over DPO-variant selection โ spend it on
getting the data shape and scale right.
- **DPO is for taste, GRPO+RLVR is for
reasoning.** Preference pairs that encode a
subjective judgment (tone, style, "which answer
is better") route to DPO. Tasks with a
verifiable pass/fail signal (math, code, tool
calls) route to GRPO+RLVR instead.
- **RL is not the fix for a model that never
succeeds.** GRPO and other RL methods sharpen
an existing capability โ they don't teach one
from zero. If the model doesn't yet understand
the task or output format, run SFT first; only
bring in RL once the model succeeds at least
sometimes.
### Common Routing Mistakes
- Reaching for fine-tuning to fix facts that
change weekly โ that's a RAG problem, and
fine-tuning will just go stale faster than the
source data does.
- Picking a DPO variant before checking whether
the actual bottleneck is data quality or model
scale โ variant choice is the ~1pp lever, not
the ~50pp one.
- Starting an RL run on a model that fails every
rollout โ route to SFT first so RL has
something to sharpen.
- Treating CPT as the default for "the model
doesn't know our domain" โ check the data
volume thresholds first; under 500MB, RAG or
RAG+fine-tune iterates faster than a CPT run.
## Model Selection
Base-model choice is size-class first, family
second, and it goes stale fast โ so it lives in
exactly one place: `references/model-catalog.md`.
That file is the only place in this plugin (and
in the DGX Spark ops plugin) that names a base
model family. Neither this skill nor
`references/memory-math.md` names one; both
describe models by size class only (for example,
"8B-class LoRA," not a model name).
The catalog is dated on purpose โ model rankings
turn over quarterly. It carries a "last verified"
date and a refresh checklist. Before trusting a
row, check that date; if stale, work the refresh
checklist in the catalog before recommending a
model from it.
**Precedence when the catalog and a method skill
disagree:** the catalog's per-row Notes column
states hardware/size-class *feasibility*, not a
method recommendation โ `lora-qlora-recipes`'s
LoRA vs QLoRA vs Full FT table (routed by task
shape) governs the actual method choice.
## Memory Feasibility
Before committing to a method, size it: total
memory โ **params ร dtype bytes + optimizer
state + gradients + activations**. Work each
term for the chosen dtype and method (full
fine-tune, LoRA, or QLoRA) โ worked worksheets
and size-class examples live in
`references/memory-math.md`.
On DGX Spark specifically, unified-memory
behavior breaks the naive estimate (transient
load peaks, `nvidia-smi` underreporting, thermal
throttling on long runs). Once the
`dgx-spark-ops` plugin is installed, defer
Spark-specific feasibility calls to its
`spark-memory-thermal-ops` skill rather than
re-deriving them here.
## Related Skills
Once this skill has picked a method, hand off to
the skill that executes it:
- `lora-qlora-recipes` โ SFT via LoRA/QLoRA
- `preference-optimization` โ DPO, ORPO, KTO
- `grpo-rlvr-training` โ GRPO with verifiable
rewards
No method is selected before the eval harness
exists โ see `eval-harness-first`.
ยฉ 2026 Flowy ยท Free and open source
Built for Claude Code ยท Not affiliated with Anthropic
