Implementation:Ollama Ollama Llama Model Arctic

Knowledge Sources	Ollama
Domains	LLM Inference, Model Architecture
Last Updated	2025-02-15 00:00 GMT

Overview

Implements the ggml computation graph builder for the Snowflake Arctic Mixture-of-Experts architecture.

Description

The llm_build_arctic constructor builds a transformer with RoPE-based positional encoding, RMS-normalized self-attention with standard Q/K/V projections, and MoE-based feed-forward layers using build_moe_ffn across all transformer blocks. The Arctic architecture uses a dense-then-sparse MoE design where each layer has both a dense FFN and a sparse MoE FFN combined via residual addition.

Usage

Enables Ollama to run Snowflake Arctic MoE models through the llama.cpp inference engine.

Code Reference

Source Location

• Repository: Ollama
• File: llama/llama.cpp/src/models/arctic.cpp
• Lines: 1-138

Signature

llm_build_arctic::llm_build_arctic(
    const llama_model & model,
    const llm_graph_params & params) : llm_graph_context(params);

Import

#include "models.h"

I/O Contract

Inputs

Name	Type	Required	Description
model	const llama_model &	Yes	Loaded model with Arctic MoE weights
params	const llm_graph_params &	Yes	Graph construction parameters

Outputs

Name	Type	Description
ggml graph	ggml_cgraph	Complete Arctic MoE computation graph

Usage Examples

// Dispatched via llama_model::build_graph():
auto builder = llm_build_arctic(model, params);

Related Pages

• Principle:Ollama_Ollama_LLM_Inference_Pipeline