TabPFN
Deliver a pinned, task-evaluated TabPFN workflow rather than treating a sklearn-like
API as proof of compatibility, permission or predictive quality. TabPFN learns a
synthetic-data prior during pretraining and conditions on a labeled training table
at inference. Its ordinary fit is not necessarily dataset-specific gradient
training; explicit fine-tuning is a distinct operation.
When not to use
- For choosing among trees, TabPFN, TabICL, sequence models and LLMs, load
ml-engineering.
- For statistical design/calibration, use data-scientist.
- For language-model SFT/LoRA/RL, use unsloth.
- TabICL has different models/APIs and terms. This skill does not operate it.
First decision and authority
Identify classification versus regression, local package versus hosted client,
checkpoint generation, data schema and permission, commercial/production scope,
selected accelerator and run destination. Do not silently choose the latest
checkpoint: code and weight terms differ by release. A download may involve
license acceptance and authentication; task authorization does not authorize
accepting terms for the user or sending private tables to a hosted client.
This skill can install packages, download weights, cache labeled tables, run GPU
work and write artifacts. Before first mutation:
Confirm the target, scope, and rollback path before acting. Read-only discovery may proceed without confirmation.
Use existing explicit authorization when it supplies these facts. Use isolated
run directories and preserve the accepted environment/artifact. Do not delete
models, change drivers or open a service as an incidental setup repair.
Guide routing
| Task |
Read |
| Understand ICL, choose checkpoints, review applicability and limits |
Model and capacity decisions |
| Install/authenticate, local versus hosted, license and offline boundaries |
Installation and artifacts |
| Prepare/split data, preserve prediction-time features, calibrate and compare |
Evaluation workflow |
| Fine-tune, reload, recover, resolve capacity/column/class failures |
Adaptation and recovery |
| Check evidence quality, exact sources and unresolved contradictions |
Source ledger |
Execution sequence
- Run the read-only doctor; record installed package versions. Metadata is not
import/CUDA compatibility. Select the version-matched official recipe.
- Record package, checkpoint, tokenizer/text preprocessing if used, code/weight
license decisions, inference ensemble/subsample settings and local/hosted scope.
- Freeze train, development/calibration and forward held-out groups. The labeled
training table is part of the inference artifact; exclude holdout labels/context.
- Use the local manifest audit to screen limits, group/exact-feature overlap,
prediction-time feature availability, label maturity and unseen classes.
It only checks declared timestamps/groups and policy; semantic data review
remains necessary. Align columns/types and preserve train-only preprocessing.
- Run a small version-matched load/fit/predict with actual device/memory and
offline/egress evidence. Never equate model ceilings to capacity of this GPU.
- Compare a simple/tree incumbent and TabPFN on identical rows, permitted
features, tuning budgets and calibration policy. Keep per-class/time/group
regressions and complete-task resource/cost results.
- Fresh-process reload must preserve checkpoint plus labeled context, columns,
transforms and configuration. Test independently scored holdout predictions.
Use the run worksheet. Complete when the requested
surface executes, original evidence is retained, fresh reload and task scoring
are verified, and all agreed quality/resource/permission constraints hold.
Otherwise report exact failed/not-run stages. Stop after three nonconverging
compatibility passes with a minimal reproducer and version inventory.
Local helper
From this skill root, Python 3.10+ and standard library only:
python3 scripts/tabpfn_check.py doctor
python3 scripts/tabpfn_check.py audit templates/task-manifest.json
python3 -m unittest discover -s scripts -p 'test_*.py'
The bundled manifest is a synthetic contract example, not a measured model run.
The auditor accepts classification/regression, three nonempty grouped forward
splits, checkpoint/package/license records and explicit row/feature policy limits.
It prints counts/codes/row numbers and a file hash, never table values or group
identities. Exit 0 clears those checks, 1 means findings/invalid input, 2 means
invalid CLI arguments. It neither loads models nor proves license approval,
true feature timestamps, calibration, GPU placement or absence of semantic leakage.
Maximum manifest size is 16 MiB. It intentionally does not audit arbitrary CSV,
pretrained-model leakage, GPU capacity or hosted payload transfers.
Evidence boundary
The offline helper tests ran. Local TabPFN loading, checkpoint authentication,
prediction, fine-tuning and GPU execution were not performed for this skill's
research. Official reports and vendor limits are source evidence, not locally
reproduced performance. Keep these distinctions in every delivery record.
1---2name: tabpfn3description: >-4 Operate and evaluate TabPFN for local tabular classification, regression and5 dataset-specific fine-tuning: checkpoint selection, weight licensing, environment6 setup, training-context leakage, calibration, resource budgeting and reproducible7 inference artifacts. Use for TabPFN package or checkpoint workflows. Do not use8 for generic model selection, operating TabICL, or language-model fine-tuning;9 use ml-engineering, the selected tool, or unsloth respectively.10license: MIT11metadata:12 research-date: "2026-10-13"13 source: https://skillmd.com/skills/magnus919/tabpfn14 upstream: "https://github.com/PriorLabs/TabPFN"15---1617# TabPFN1819Deliver a pinned, task-evaluated TabPFN workflow rather than treating a sklearn-like20API as proof of compatibility, permission or predictive quality. TabPFN learns a21synthetic-data prior during pretraining and conditions on a labeled training table22at inference. Its ordinary `fit` is not necessarily dataset-specific gradient23training; explicit fine-tuning is a distinct operation.2425## When not to use2627- For choosing among trees, TabPFN, TabICL, sequence models and LLMs, load28 [ml-engineering](../ml-engineering/SKILL.md).29- For statistical design/calibration, use [data-scientist](../data-scientist/SKILL.md).30- For language-model SFT/LoRA/RL, use [unsloth](../unsloth/SKILL.md).31- TabICL has different models/APIs and terms. This skill does not operate it.3233## First decision and authority3435Identify classification versus regression, local package versus hosted client,36checkpoint generation, data schema and permission, commercial/production scope,37selected accelerator and run destination. Do not silently choose the latest38checkpoint: code and weight terms differ by release. A download may involve39license acceptance and authentication; task authorization does not authorize40accepting terms for the user or sending private tables to a hosted client.4142This skill can install packages, download weights, cache labeled tables, run GPU43work and write artifacts. Before first mutation:4445> Confirm the target, scope, and rollback path before acting. Read-only discovery may proceed without confirmation.4647Use existing explicit authorization when it supplies these facts. Use isolated48run directories and preserve the accepted environment/artifact. Do not delete49models, change drivers or open a service as an incidental setup repair.5051## Guide routing5253| Task | Read |54|---|---|55| Understand ICL, choose checkpoints, review applicability and limits | [Model and capacity decisions](references/models.md) |56| Install/authenticate, local versus hosted, license and offline boundaries | [Installation and artifacts](references/operations.md) |57| Prepare/split data, preserve prediction-time features, calibrate and compare | [Evaluation workflow](references/evaluation.md) |58| Fine-tune, reload, recover, resolve capacity/column/class failures | [Adaptation and recovery](references/adaptation.md) |59| Check evidence quality, exact sources and unresolved contradictions | [Source ledger](references/sources.md) |6061## Execution sequence62631. Run the read-only doctor; record installed package versions. Metadata is not64 import/CUDA compatibility. Select the version-matched official recipe.652. Record package, checkpoint, tokenizer/text preprocessing if used, code/weight66 license decisions, inference ensemble/subsample settings and local/hosted scope.673. Freeze train, development/calibration and forward held-out groups. The labeled68 training table is part of the inference artifact; exclude holdout labels/context.694. Use the local manifest audit to screen limits, group/exact-feature overlap,70 prediction-time feature availability, label maturity and unseen classes.71 It only checks declared timestamps/groups and policy; semantic data review72 remains necessary. Align columns/types and preserve train-only preprocessing.735. Run a small version-matched load/fit/predict with actual device/memory and74 offline/egress evidence. Never equate model ceilings to capacity of this GPU.756. Compare a simple/tree incumbent and TabPFN on identical rows, permitted76 features, tuning budgets and calibration policy. Keep per-class/time/group77 regressions and complete-task resource/cost results.787. Fresh-process reload must preserve checkpoint plus labeled context, columns,79 transforms and configuration. Test independently scored holdout predictions.8081Use [the run worksheet](templates/run-record.md). Complete when the requested82surface executes, original evidence is retained, fresh reload and task scoring83are verified, and all agreed quality/resource/permission constraints hold.84Otherwise report exact failed/not-run stages. Stop after three nonconverging85compatibility passes with a minimal reproducer and version inventory.8687## Local helper8889From this skill root, Python 3.10+ and standard library only:9091```bash92python3 scripts/tabpfn_check.py doctor93python3 scripts/tabpfn_check.py audit templates/task-manifest.json94python3 -m unittest discover -s scripts -p 'test_*.py'95```9697The bundled manifest is a synthetic contract example, not a measured model run.98The auditor accepts classification/regression, three nonempty grouped forward99splits, checkpoint/package/license records and explicit row/feature policy limits.100It prints counts/codes/row numbers and a file hash, never table values or group101identities. Exit 0 clears those checks, 1 means findings/invalid input, 2 means102invalid CLI arguments. It neither loads models nor proves license approval,103true feature timestamps, calibration, GPU placement or absence of semantic leakage.104Maximum manifest size is 16 MiB. It intentionally does not audit arbitrary CSV,105pretrained-model leakage, GPU capacity or hosted payload transfers.106107## Evidence boundary108109The offline helper tests ran. Local TabPFN loading, checkpoint authentication,110prediction, fine-tuning and GPU execution were not performed for this skill's111research. Official reports and vendor limits are source evidence, not locally112reproduced performance. Keep these distinctions in every delivery record.