Implementation:Ollama Ollama XCreate Client

Knowledge Sources	Ollama
Domains	Model Creation, Safetensors
Last Updated	2025-02-15 00:00 GMT

Overview

Provides client-side model creation for importing safetensors-based models (both LLM and image generation) directly to disk, bypassing the HTTP API.

Description

CreateModel detects whether a directory contains a safetensors LLM model (config.json + .safetensors) or image generation model (model_index.json), then delegates to the appropriate creation function. Creates blob layers on disk via manifest.NewLayer, supports optional quantization (int4, int8, nvfp4, mxfp8) through callback-based architecture to avoid import cycles. Writes manifests with model capabilities, parser/renderer names, and Modelfile configurations (template, system prompt, license).

Usage

Called by ollama create when importing models from local directories in the experimental safetensors format.

Code Reference

Source Location

• Repository: Ollama
• File: x/create/client/create.go
• Lines: 1-480

Signature

type ModelfileConfig struct {
    Template string
    System   string
    License  string
}

type CreateOptions struct {
    ModelName string
    ModelDir  string
    Quantize  string
    Modelfile *ModelfileConfig
}

func CreateModel(opts CreateOptions, p *progress.Progress) error

Import

import "github.com/ollama/ollama/x/create/client"

I/O Contract

Inputs

Name	Type	Required	Description
opts	CreateOptions	Yes	Model name, directory path, quantization type, and Modelfile config
p	*progress.Progress	Yes	Progress reporter for UI feedback

Outputs

Name	Type	Description
error	error	Non-nil if model creation fails

Usage Examples

p := progress.NewProgress(os.Stderr)
err := client.CreateModel(client.CreateOptions{
    ModelName: "my-model:latest",
    ModelDir:  "/path/to/model",
    Quantize:  "int4",
    Modelfile: &client.ModelfileConfig{
        Template: "chatml",
        System:   "You are a helpful assistant.",
    },
}, p)

Related Pages

• Principle:Ollama_Ollama_ModelCreation