Implementation:Ollama Ollama ConvertAdapter

Knowledge Sources	Ollama
Domains	Model_Architecture, Fine_Tuning
Last Updated	2026-02-14 00:00 GMT

Overview

Concrete tool for converting LoRA adapters from SafeTensors to GGUF format provided by the convert package.

Description

ConvertAdapter reads a LoRA adapter directory (containing adapter_config.json and safetensors files), detects the adapter architecture (LLaMA, Gemma2, etc.), converts tensor names and layouts to GGUF format, and writes the result as a GGUF file.

Architecture-specific converters like llamaAdapter handle name remapping (e.g., model.layers.*.self_attn.q_proj.lora_A to blk.*.attn_q.weight.loraA) and weight repacking (Q/K head interleaving for LLaMA models).

Usage

Called by CreateHandler when processing an ADAPTER directive during model creation.

Code Reference

Source Location

• Repository: ollama
• File: convert/convert.go (ConvertAdapter), convert/convert_llama_adapter.go (llamaAdapter)
• Lines: convert.go:L217-253 (ConvertAdapter), convert_llama_adapter.go:L22-170 (llamaAdapter)

Signature

func ConvertAdapter(fsys fs.FS, f *os.File, baseKV ofs.Config) error

func (p *llamaAdapter) KV(baseKV fs.Config) KV
func (p *llamaAdapter) Tensors(ts []Tensor) []*ggml.Tensor
func (p *llamaAdapter) Replacements() []string

Import

import "github.com/ollama/ollama/convert"

I/O Contract

Inputs

Name	Type	Required	Description
fsys	fs.FS	Yes	Filesystem rooted at adapter directory (with adapter_config.json and .safetensors)
f	*os.File	Yes	Output file for GGUF-formatted adapter
baseKV	ofs.Config	Yes	Base model configuration (architecture, dimensions) needed for correct mapping

Outputs

Name	Type	Description
error	error	Non-nil if adapter format is unsupported or conversion fails
Side effect	GGUF file	Converted adapter written to output file

Usage Examples

Internal Usage in CreateHandler

// From server/create.go
adapterFS := os.DirFS(adapterPath)
outFile, _ := os.CreateTemp("", "adapter-*.gguf")
err := convert.ConvertAdapter(adapterFS, outFile, baseModelConfig)
if err != nil {
    return err
}
// Use the GGUF adapter file in the model manifest

Related Pages

Implements Principle

• Principle:Ollama_Ollama_Adapter_Conversion

Requires Environment

• Environment:Ollama_Ollama_Go_Runtime