Implementation:Ollama Ollama Llama Model SmolLM3

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 SmolLM3 model architecture.

Description

The llm_build_smollm3 constructor builds a transformer with conditional RoPE application -- some layers skip RoPE based on n_no_rope_layer_step. Features configurable attention scaling via hparams.f_attention_scale, RMS-normalized self-attention with Q/K/V projections and optional biases, and SiLU-gated feed-forward layers across all transformer blocks.

Usage

Enables Ollama to run SmolLM3 models through the llama.cpp inference engine, supporting its unique architecture feature where certain layers operate without rotary position embeddings.

Code Reference

Source Location

• Repository: Ollama
• File: llama/llama.cpp/src/models/smollm3.cpp
• Lines: 1-128

Signature

llm_build_smollm3::llm_build_smollm3(
    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 SmolLM3 weights
params	const llm_graph_params &	Yes	Graph construction parameters

Outputs

Name	Type	Description
ggml graph	ggml_cgraph	Complete SmolLM3 computation graph with conditional RoPE

Usage Examples

auto builder = llm_build_smollm3(model, params);

Related Pages

• Principle:Ollama_Ollama_LLM_Inference_Pipeline