✨ feat(tasks): add LlamaFarm to local-apps

[rgthelen]

PR Description

Summary

Adds LlamaFarm to the local apps list, enabling HuggingFace users to run models locally with LlamaFarm's config-based ML workflows.

What is LlamaFarm?

LlamaFarm is an open-source AI platform that runs entirely on local hardware. It combines model inference with specialized ML capabilities — all in one local runtime.

Supported Model Categories

Category	Examples
Text Generation	Llama3, Mistral, Qwen3 Phi, GPT-OSS
Embeddings	sentence-transformers, BGE, nomic-embed
Audio	Whisper, Wav2Vec2
Document/OCR	LayoutLM, Donut, Surya

Built-in ML Capabilities (beyond model inference)

Capability	Description
Text Classification	Train custom classifiers with 8-16 examples (SetFit)
Anomaly Detection	Isolation Forest, One-Class SVM, LOF, Autoencoders
Named Entity Recognition	Extract people, organizations, locations
OCR & Document Extraction	Surya, EasyOCR, PaddleOCR, Tesseract
Reranking	Cross-encoder models for improved RAG quality
RAG Pipeline	Full ingestion, chunking, retrieval, and querying

Key features:

• 🔒 Complete privacy — data never leaves the device
• 💰 No API costs — uses open-source models
• 🌐 Offline capable — works without internet after model download
• ⚡ Hardware optimized — automatic GPU/NPU acceleration (Apple Silicon, NVIDIA, AMD)
• 📦 GGUF support — runs quantized models via llama.cpp
• 🧠 Built-in ML — anomaly detection, classifiers, NER without external services

Installation Methods

• CLI: curl -fsSL https://raw.githubusercontent.com/llama-farm/llamafarm/main/install.sh | bash
• Desktop App: Available for macOS, Windows, and Linux

Links

• Website: https://llamafarm.dev
• Docs: https://docs.llamafarm.dev
• GitHub: https://github.com/llama-farm/llamafarm
• Discord: https://discord.gg/RrAUXTCVNF

---

Code Changes

File: packages/tasks/src/local-apps.ts

1. Add snippet function (after snippetLemonade):

const snippetLlamaFarm = (model: ModelData, filepath?: string): LocalAppSnippet[] => {
	const modelId = model.id;
	const isGguf = isLlamaCppGgufModel(model);
	const tagName = isGguf ? getQuantTag(filepath) : "";

	// Determine model type hint for better UX
	const getModelTypeHint = (): string => {
		if (model.pipeline_tag === "automatic-speech-recognition") {
			return "# Transcribe audio:";
		}
		if (model.pipeline_tag === "feature-extraction" || model.pipeline_tag === "sentence-similarity") {
			return "# Generate embeddings:";
		}
		return "# Chat with this model:";
	};

	return [
		{
			title: "Install LlamaFarm CLI (config-based ML workflows)",
			setup: [
				"# macOS / Linux:",
				"curl -fsSL https://raw.githubusercontent.com/llama-farm/llamafarm/main/install.sh | bash",
				"",
				"# Windows (PowerShell):",
				"# irm https://raw.githubusercontent.com/llama-farm/llamafarm/main/install.ps1 | iex",
			].join("\n"),
			content: [
				"# Initialize a project:",
				"lf init my-project",
				"lf start",
				"",
				getModelTypeHint(),
				`lf chat --model ${modelId}${tagName} "Hello!"`,
				"",
				"# Or open the visual designer:",
				"# http://localhost:8000",
			].join("\n"),
		},
		{
			title: "Or download the Desktop App (no CLI required)",
			setup: "# Download from GitHub releases:",
			content: [
				"# macOS (Universal):",
				"# https://github.com/llama-farm/llamafarm/releases/latest/download/LlamaFarm-desktop-app-mac-universal.dmg",
				"#",
				"# Windows:",
				"# https://github.com/llama-farm/llamafarm/releases/latest/download/LlamaFarm-desktop-app-windows.exe",
				"#",
				"# Linux (x86_64):",
				"# https://github.com/llama-farm/llamafarm/releases/latest/download/LlamaFarm-desktop-app-linux-x86_64.AppImage",
				"#",
				"# Linux (ARM64):",
				"# https://github.com/llama-farm/llamafarm/releases/latest/download/LlamaFarm-desktop-app-linux-arm64.AppImage",
			].join("\n"),
		},
		{
			title: "Built-in ML capabilities (beyond model inference)",
			content: [
				"# In addition to inference, LlamaFarm includes specialized ML features:",
				"#",
				"# • Text Classification — Train classifiers with 8-16 examples (SetFit)",
				"# • Anomaly Detection — Isolation Forest, One-Class SVM, LOF, Autoencoders",
				"# • Named Entity Recognition — Extract people, orgs, locations",
				"# • OCR & Document Extraction — Surya, EasyOCR, PaddleOCR",
				"# • Reranking — Cross-encoders for better RAG retrieval",
				"# • Full RAG Pipeline — Ingest PDFs/docs, chunk, embed, query",
				"#",
				"# See: https://docs.llamafarm.dev/docs/models",
			].join("\n"),
		},
	];
};

2. Add to LOCAL_APPS object (alphabetically, after lemonade):

llamafarm: {
	prettyLabel: "LlamaFarm",
	docsUrl: "https://llamafarm.dev",
	mainTask: "text-generation",
	displayOnModelPage: (model) =>
		// Text generation (GGUF, Transformers, TGI)
		isLlamaCppGgufModel(model) ||
		isTransformersModel(model) ||
		model.pipeline_tag === "text-generation" ||
		// Embeddings
		model.pipeline_tag === "feature-extraction" ||
		model.pipeline_tag === "sentence-similarity" ||
		// Audio models
		model.pipeline_tag === "automatic-speech-recognition" ||
		model.pipeline_tag === "audio-classification" ||
		// NLP tasks (NER, classification)
		model.pipeline_tag === "token-classification" ||
		model.pipeline_tag === "text-classification" ||
		// Document understanding
		model.pipeline_tag === "document-question-answering",
	snippet: snippetLlamaFarm,
},

---

Testing

Tested snippet generation for various model types:

• GGUF models (e.g., qwen/qwen3-8b-gguf)
• Transformers text models (e.g., microsoft/phi-2)
• Embedding models (e.g., sentence-transformers/all-MiniLM-L6-v2)
• ASR models (e.g., openai/whisper-large-v3)

---

Checklist

• Code follows the existing patterns in local-apps.ts
• Snippet function returns LocalAppSnippet[]
• displayOnModelPage covers all supported model types
• Documentation links are valid
• No breaking changes to existing apps

[rgthelen]

The test failures are due to test issues, not this PR. Please review #1955; it fixes the test failures (the failures were due to a vLLM task update, but the tests were not updated, and some stale API specs).