New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Select Model

by PINA-org in PINA-org/PINA

Guides users through selecting, configuring, or creating a PyTorch model for a PINA problem.

MITAuto-check passedAI & LLM Engineering

Install Select Model

skills CLI
$ npx skills add PINA-org/PINA --skill select-model -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install PINA-org/PINA select-model --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/PINA-org/PINA.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.opencode/skills/select-model .claude/skills/select-model && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
select-model
GitHub stars
797
Token cost
~2.4k tokens
SKILL.md length
893 words
Files
1
Skills in repo
10
Repo updated
First seen
Licence
MIT

At a glance

Guides users through selecting, configuring, or creating a PyTorch model for a PINA problem.

  • Works in 4 steps: Understand the problem together → Reason about architecture → Enhancements and embeddings → …
  • The user asks about models
  • SKILL.md covers Step 1 — Understand the…, Step 2 — Reason about…, Step 3 — Enhancements and… and Step 4 — Build and test, plus 1 more section
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Select Model is an agent skill from PINA-org/PINA. Guides users through selecting, configuring, or creating a PyTorch model for a PINA problem. Use when the user asks about models, architecture, neural network choice, or how to build/choose a model for their PINA problem. Also triggers on phrases like "what model should I use", "choose a model", "create a model", "neural network", "architecture", "model selection", or when the user mentions specific model names (FeedForward, DeepONet, FNO, PirateNet, etc.).

Its SKILL.md is about 2.4k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts. Compatibility notes: opencode, codex, claude

It sits in AI & LLM Engineering, covering Deep learning. It works with PyTorch. The repository describes itself as: Physics-Informed Neural networks for Advanced modeling. The licence is MIT.

When your agent uses it

  • The user asks about models
  • Neural network choice
  • How to build/choose a model for their PINA problem
  • Phrases like what model should I use

Example prompts

  • “what model should I use”
  • “choose a model”
  • “create a model”
  • “/select-model”

Requirements

  • Python 3
  • Compatibility (from SKILL.md): opencode, codex, claude

Workflow steps

4 steps, taken from the step headings in SKILL.md.

  1. Understand the problem together
  2. Reason about architecture
  3. Enhancements and embeddings
  4. Build and test

What it can do on your machine

Read from SKILL.md and the folder at commit 0cd8afb. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md (its code samples are python).

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

  • Compatibility

    opencode, codex, claude

    From compatibility in the SKILL.md frontmatter.

Context cost

Select Model loads about 2.4k tokens when it runs. Until then it costs about 119 tokens; SKILL.md has 893 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~119
When it runs · the whole SKILL.md, loaded when a task matches
~2.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from PINA-org/PINA at commit 0cd8afb, republished under its MIT licence (© PINA-org). 893 words, ~2,422 tokens.

Download SKILL.mdSave it as .claude/skills/select-model/SKILL.md (or your agent's skills folder).
name
select-model
description
Guides users through selecting, configuring, or creating a PyTorch model for a PINA problem. Use when the user asks about models, architecture, neural network choice, or how to build/choose a model for their PINA problem. Also triggers on phrases like "what model should I use", "choose a model", "create a model", "neural network", "architecture", "model selection", or when the user mentions specific model names (FeedForward, DeepONet, FNO, PirateNet, etc.).
compatibility
opencode, codex, claude
license
MIT
metadata.audience
users
metadata.workflow
model-creation

Select a Model for a PINA Problem

[!IMPORTANT] Read RULES.md before using this skill — it applies to all skills.

Use this skill to choose the right neural network architecture for a PINA problem, configure its parameters, and optionally enhance it with embeddings or blocks.

Step 1 — Understand the problem together

Start by gathering enough context to make a reasoned recommendation. Ask the user these questions conversationally — one at a time, adapting to their answers — rather than firing them all at once.

  • Input/output: What variables go in and come out? (e.g., input: x, y, t → output: u, v). If they have a PINA problem object, extract directly: len(problem.input_variables) / len(problem.output_variables).
  • Data shape: What does a single sample look like? Scalar values per point? A field on a grid? A graph? Sensor readings + coordinates?
  • Data structure (determines what operations fit): A flat table of points (linear / MLP layers), a regular grid (conv1d/conv2d/conv3d, or FNO), an irregular mesh or point cloud (message passing / graph nets, or coordinate-based MLPs like PINNs), or sequences / time series (RNN / transformer)?
  • Problem type: Are you solving a PDE (physics-informed), learning an operator (map function to function), fitting a curve/surface, or discovering equations from data?
  • Constraints: Roughly how much data? Any compute or time limits? Is accuracy-critical or is a rough baseline enough?

If the user cannot describe their problem at all, reply:

I need at least the input/output dimensions to suggest an architecture. What are the input variables and what are the outputs you want to predict?

Step 2 — Reason about architecture

Take the context from Step 1 and reason about which architecture family fits best. Present the reasoning to the user and invite their input rather than simply prescribing.

Available building blocks

Models live in PINA's model (pina.model), with smaller blocks in pina.model.block. Browsing these modules is the best way to discover what's available — the sections below discuss how to choose.

Reasoning guide

Work through these considerations in dialogue with the user:

  1. Operator learning? — mapping between function spaces? That points to FNO (regular grid), DeepONet/MIONet (sensor→field), or neural operators (Averaging/LowRank/Graph). Ask about the data geometry.

  2. Message-passing / GNN? — input is a graph with explicit connectivity (mesh, point cloud, network)?

    • PINA has many GNN based blocks in pina.model.block.message_passing.

  3. Equation discovery? — SINDy or similar. Make sure the user knows this discovers symbolic expressions, not just a black-box surrogate.

  4. Spline fitting? — Spline for 1D, SplineSurface for 2D, or VectorizedSpline for independent per-feature splines. Typically explicit fitting, not deep learning.

  5. Standard pointwise mapping? — This is where most PDE problems land. Within this family, consider:

    • Complexity: Start simple (FeedForward). Only escalate if it struggles — PirateNet for multi-scale, ResidualFeedForward for depth.
    • Data regime: Lots of collocation points → deeper/larger MLP works. Very little data → KAN can be parameter-efficient.
    • Multi-scale / high-frequency: FourierFeatureEmbedding (Step 3) is a cheap add-on to any MLP. PirateNet bundles it natively.

  6. Custom nn.Module — When none of the above fit, use pure PyTorch. This is common for:

    • Multi-input/multi-output architectures PINA doesn't have
    • Pretrained encoders (CNNs, transformers)
    • Novel research architectures
    • Lightweight models where PINA's wrapping adds unnecessary overhead

    The only requirement: it must be a valid torch.nn.Module that accepts a tensor/graph and returns a tensor/graph. With PINA solvers, keep the constructor parameter names input_dimensions and output_dimensions so labels are handled automatically (or use nn.Module directly without PINA).

Show full SKILL.md (345 more words)Show less

Step 3 — Enhancements and embeddings

These are add-ons to improve convergence or enforce structure. Discuss whether the user's problem benefits from them.

Fourier Feature Embedding (multi-scale)

Use when the PDE solution has features at multiple length scales, or when a plain MLP converges slowly (spectral bias).

python
from pina.model.block import FourierFeatureEmbedding

embedding = FourierFeatureEmbedding(
    input_dimension=input_dim,
    output_dimension=128,  # must be even
    sigma=2.0,
)

Wrap any model manually:

python
class WithEmbedding(nn.Module):
    def __init__(self, input_dim, output_dim, inner_size=64):
        super().__init__()
        self.embedding = FourierFeatureEmbedding(input_dim, 128)
        self.net = nn.Sequential(
            nn.Linear(128, inner_size), nn.Tanh(),
            nn.Linear(inner_size, output_dim),
        )

    def forward(self, x):
        return self.net(self.embedding(x))
Periodic Boundary Embedding

Hard-code periodicity so the model never needs to learn it:

python
from pina.model.block import PeriodicBoundaryEmbedding

embedding = PeriodicBoundaryEmbedding(
    input_dimension=2,
    periods={"x": 2 * 3.14159},
)
Residual / Orthogonal blocks
  • ResidualBlock(input_dim, output_dim, hidden_dim, activation) — stacks layers with a skip connection. Use inside a custom model.
  • OrthogonalBlock(dim=-1) — orthonormalize features along a dimension. Helps training stability.

Both are in pina.model.block.

Message-passing blocks (graph models)

For graph-structured inputs, PINA provides blocks that follow the same pattern: update node features by aggregating neighbor information. Available in pina.model.block (top-level) and pina.model.block.message_passing:

BlockSourceDescription
InteractionNetworkBlock.message_passingStandard encode–process–decode interaction network
DeepTensorNetworkBlock.message_passingHigher-order tensor message passing
EnEquivariantNetworkBlock.message_passingEnergy-conserving equivariant message passing
EquivariantGraphNeuralOperatorBlock.message_passingSE(3)-equivariant neural operator block
RadialFieldNetworkBlock.message_passingRadial-basis field convolution on point clouds
GNOBlock.block (top-level)Graph neural operator kernel integration

Prefer the built-in GraphNeuralOperator (in pina.model) if it matches your problem — it composes lifting, message passing, and projection into one class. Drop down to these blocks when you need a custom arrangement.

Step 4 — Build and test

Construction pattern

Prefer PINA's built-in models when they match (less boilerplate). Use pure nn.Module when they don't.

python
# PINA model
from pina.model import FeedForward
model = FeedForward(input_dimensions=3, output_dimensions=3, inner_size=64, n_layers=4, func=nn.Tanh)

# Pure torch custom model
class MyModel(nn.Module):
    def __init__(self, input_dimensions, output_dimensions):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(input_dimensions, 128), nn.Tanh(),
            nn.Linear(128, output_dimensions),
        )
    def forward(self, x):
        return self.net(x)

model = MyModel(input_dimensions=3, output_dimensions=3)

# Graph model: lifting → message passing → projection
from pina.model.block import GNOBlock, FeedForward

class GraphModel(nn.Module):
    def __init__(self, input_dim, hidden_dim, output_dim):
        super().__init__()
        self.lifting = FeedForward(input_dim, hidden_dim)
        self.gno = GNOBlock(...)
        self.projection = FeedForward(hidden_dim, output_dim)

    def forward(self, x, edge_index, edge_attr):
        x = self.lifting(x)
        x = self.gno(x, edge_index, edge_attr)
        return self.projection(x)
Validation
  • Forward pass — tensor model: model(torch.randn(batch, input_dim)) → (batch, output_dim).
  • Forward pass — graph model: model(x, edge_index, edge_attr) with dummy tensors of the right shapes, or a torch_geometric.data.Data object if the model expects one. The output should match the expected node-feature shape.
  • If using PINA solvers: confirm the solver can call model.forward() with the actual condition tensors (same dtype, device, shapes).
  • If using custom nn.Module with PINA: keep input_dimensions and output_dimensions as constructor kwargs for automatic label extraction.

Checklist

  • Input/output dimensions determined (from problem object or user Q&A)
  • Architecture reasoning discussed with the user, not prescribed
  • Model class selected (PINA built-in, or custom nn.Module)
  • Embeddings/enhancements added if the problem needs them (multi-scale, periodic boundaries, residual connections)
  • Model passes a forward and backward pass smoke test with correct shapes

© PINA-org, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .opencode/skills/select-model of PINA-org/PINA.

Open the folder on GitHubat commit 0cd8afb

Compare with similar skills

Select Model next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Select Model compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Select Model this skillPINA-org/PINA797—~2.4kAutomated safety check: PassMIT
Add Uint Supportpytorch/pytorch104k2 repos~2.3kAutomated safety check: PassCustom licence
CLIP Image-Text MatchingOrchestra-Research/AI-Research-SKILLs13k8 repos~1.7kAutomated safety check: PassMIT
Add Torch Shapes Examplefacebook/pyrefly7.1k—~1.3kAutomated safety check: PassMIT
Interview Cheatsheetwanshuiyin/ARIS-in-AI-Offer5801 repos~3.4kAutomated safety check: NotesMIT
Ghstack CIpytorch/pytorch104k—~1.4kAutomated safety check: PassCustom licence

Similar skills

  • Add Uint Support

    pytorch/pytorch

    Add unsigned integer (uint) type support to PyTorch operators by updating ATDISPATCH macros.

    104k GitHub starsUsed in 2 repos~2.3k tokens
    AI & LLM EngineeringAuto-check passed

  • CLIP Image-Text Matching

    Orchestra-Research/AI-Research-SKILLs

    Explains OpenAI's CLIP model for zero-shot image classification, image-text similarity, semantic image search and content moderation, with install steps and code patterns.

    13k GitHub starsUsed in 8 repos~1.7k tokens
    AI & LLM EngineeringAuto-check passed

  • Add Torch Shapes Example

    facebook/pyrefly

    Official

    A skill your agent uses when adding a new PyTorch model to Pyrefly's shape-tracking example corpus under tensor-shapes/pyrefly-torch-stubs/examples — i.e.

    7.1k GitHub stars~1.3k tokensUpdated today
    AI & LLM EngineeringAuto-check passed

  • Interview Cheatsheet

    wanshuiyin/ARIS-in-AI-Offer

    Generate a long-form Chinese interview-prep cheat sheet on a specific ML/LLM topic — formulas with derivations, from-scratch PyTorch code, comparison tables, and 25 高频面试题 (L1 必会 / L2 进阶 / L3 顶级 lab).

    580 GitHub starsUsed in 1 repo~3.4k tokens
    AI & LLM EngineeringAuto-check: notes

  • Ghstack CI

    pytorch/pytorch

    Manage CI for PyTorch ghstack stacks by running CI where its results are useful now and deferring other PRs with [no-ci].

    104k GitHub stars~1.4k tokensUpdated today
    AI & LLM EngineeringAuto-check passed

  • MUSA GPU Training Optimizer

    open-infra-skills/infra-skills

    Profiles, benchmarks and tunes AI training workloads on Moore Threads MUSA GPUs with a measurement-first process that keeps model behavior unchanged.

    141 GitHub stars~1.7k tokensUpdated 3 mo ago
    AI & LLM EngineeringAuto-check passed

More from PINA-org/PINA

All 10 skills in this repo

  • Create Skill

    PINA-org/PINA

    Create, modify, and improve PINA skills. An agent skill from PINA-org/PINA.

    797 GitHub stars~1.8k tokensUpdated yesterday
    Auto-check passed

  • Pina Workflow

    PINA-org/PINA

    Orchestrates a complete session from problem definition through trained solver.

    797 GitHub stars~1.1k tokensUpdated yesterday
    Auto-check passed

  • Select Solver

    PINA-org/PINA

    Guides users through selecting the right PINA solver for their problem, or creating a custom solver when no built-in fits.

    797 GitHub stars~2.9k tokensUpdated yesterday
    Auto-check passed

  • Select Trainer

    PINA-org/PINA

    Guides users through configuring the PINA Trainer for solver training, including batching strategy, data splitting, and Lightning options.

    797 GitHub stars~1.3k tokensUpdated yesterday
    Auto-check passed

  • Skill Sync Checker

    PINA-org/PINA

    Audits SKILL.md files in this repo's skills directory for references to functions, classes, or modules (mentioned by name in prose, e.g.

    797 GitHub stars~1.5k tokensUpdated yesterday
    Auto-check passed

  • Condition Setup

    PINA-org/PINA

    Set up conditions for PINA problems. An agent skill from PINA-org/PINA.

    797 GitHub stars~1k tokensUpdated yesterday
    Auto-check passed

Works with

Categories

Questions about Select Model

What does Select Model do?

Guides users through selecting, configuring, or creating a PyTorch model for a PINA problem. Select Model is an agent skill from PINA-org/PINA. Guides users through selecting, configuring, or creating a PyTorch model for a PINA problem.

When should I use Select Model?

Select Model fits situations like: the user asks about models; neural network choice; how to build/choose a model for their PINA problem; phrases like what model should I use.

How do I install Select Model in Claude Code?

Run `npx skills add PINA-org/PINA --skill select-model -a claude-code`. Or copy the skill folder (.opencode/skills/select-model in PINA-org/PINA) into .claude/skills/select-model in your project. Claude Code loads it when a task matches its description.

How do I install Select Model in Codex?

Run `npx skills add PINA-org/PINA --skill select-model -a codex`. Or copy the skill folder (.opencode/skills/select-model in PINA-org/PINA) into .agents/skills/select-model in your project. Codex loads it when a task matches its description.

Can I use Select Model in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add PINA-org/PINA --skill select-model -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/select-model, .gemini/skills/select-model, .github/skills/select-model and .opencode/skills/select-model in your project.

What does Select Model need to run?

SKILL.md names no scripts, command-line tools or credentials: Select Model is instructions for the agent only. Our summary lists: Python 3. Compatibility (from SKILL.md): opencode, codex, claude.

Does Select Model access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Select Model safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Select Model use?

Select Model is published under the MIT licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Select Model use?

About 2.4k tokens (SKILL.md is roughly 9.7k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Select Model?

Skills that share tags, products or a category with Select Model: Add Uint Support (pytorch/pytorch, 104k stars), CLIP Image-Text Matching (Orchestra-Research/AI-Research-SKILLs, 13k stars), Add Torch Shapes Example (facebook/pyrefly, 7.1k stars) and Interview Cheatsheet (wanshuiyin/ARIS-in-AI-Offer, 580 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Select Model?

PINA-org (a GitHub organization) maintains it in PINA-org/PINA, which has 797 GitHub stars. The repository holds 10 skills in this directory. The repository was last updated on October 6, 2026.

Source: PINA-org/PINA on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.