Implementation:Ollama Ollama Llama Model Apertus

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 Apertus model architecture.

Description

The llm_build_apertus constructor builds a standard transformer pipeline: token embeddings, RoPE-based positional encoding with rope factors, RMS-normalized self-attention with Q/K/V projections (including Q and K normalization), and SiLU-gated feed-forward layers across all transformer blocks, producing final logits via an output projection.

Usage

Enables Ollama to run Apertus-family models through the llama.cpp inference engine by defining how model weights map to ggml tensor operations.

Code Reference

Source Location

• Repository: Ollama
• File: llama/llama.cpp/src/models/apertus.cpp
• Lines: 1-125

Signature

llm_build_apertus::llm_build_apertus(
    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 Apertus weights
params	const llm_graph_params &	Yes	Graph construction parameters

Outputs

Name	Type	Description
ggml graph	ggml_cgraph	Complete Apertus computation graph

Usage Examples

// Dispatched via llama_model::build_graph():
auto builder = llm_build_apertus(model, params);
// Graph is constructed in the constructor

Related Pages

• Principle:Ollama_Ollama_LLM_Inference_Pipeline