Safetensors Model

Introduction

Burn supports importing model weights from the Safetensors format, a secure and efficient alternative to pickle-based formats. Whether you've trained your model in PyTorch or you want to use a pre-trained model that provides weights in Safetensors format, you can easily import them into Burn.

This guide demonstrates the complete workflow for exporting models to Safetensors format and importing them into Burn.

Exporting Models to Safetensors Format

To export a PyTorch model to Safetensors format, you'll need the safetensors Python library. This library provides a simple API for saving model weights in the Safetensors format.

Example: Exporting a PyTorch Model

import torch
import torch.nn as nn
from safetensors.torch import save_file

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(2, 2, (2,2))
        self.conv2 = nn.Conv2d(2, 2, (2,2), bias=False)

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        return x

if __name__ == "__main__":
    # Set seed for reproducibility
    torch.manual_seed(42)

    # Initialize model and ensure it's on CPU
    model = Net().to(torch.device("cpu"))

    # Extract model weights dictionary
    model_weights = model.state_dict()

    # Save to Safetensors format
    save_file(model_weights, "conv2d.safetensors")

Verifying the Export

You can verify your exported model by viewing the .safetensors file in Netron, a neural network visualization tool. A correctly exported file will display a flat structure of tensors, similar to a PyTorch .pt weights file.

Importing Safetensors Models into Burn

Importing a Safetensors model into Burn involves two main steps:

1. Defining the model architecture in Burn
2. Loading the weights from the Safetensors file

Step 1: Define the Model in Burn

First, you need to create a Burn model that matches the architecture of the model you exported:

#![allow(unused)]
fn main() {
use burn::{
    nn::conv::{Conv2d, Conv2dConfig},
    prelude::*,
};

#[derive(Module, Debug)]
pub struct Net<B: Backend> {
    conv1: Conv2d<B>,
    conv2: Conv2d<B>,
}

impl<B: Backend> Net<B> {
    /// Create a new model.
    pub fn init(device: &B::Device) -> Self {
        let conv1 = Conv2dConfig::new([2, 2], [2, 2])
            .init(device);
        let conv2 = Conv2dConfig::new([2, 2], [2, 2])
            .with_bias(false)
            .init(device);
        Self { conv1, conv2 }
    }

    /// Forward pass of the model.
    pub fn forward(&self, x: Tensor<B, 4>) -> Tensor<B, 4> {
        let x = self.conv1.forward(x);
        self.conv2.forward(x)
    }
}
}

Step 2: Load the Weights

You have two options for loading the weights:

Option A: Load Dynamically at Runtime

This approach loads the Safetensors file directly at runtime, requiring the burn-import dependency:

use crate::model;
use burn::record::{FullPrecisionSettings, Recorder};
use burn_import::safetensors::SafetensorsFileRecorder;

type Backend = burn_ndarray::NdArray<f32>;

fn main() {
    let device = Default::default();

    // Load weights from Safetensors file
    let record = SafetensorsFileRecorder::<FullPrecisionSettings>::default()
        .load("./conv2d.safetensors".into(), &device)
        .expect("Should decode state successfully");

    // Initialize model and load weights
    let model = model::Net::<Backend>::init(&device).load_record(record);
}

Option B: Pre-convert to Burn's Binary Format

This approach converts the Safetensors file to Burn's optimized binary format during build time, removing the runtime dependency on burn-import:

#![allow(unused)]
fn main() {
// This code would go in build.rs or a separate tool

use crate::model;
use burn::record::{FullPrecisionSettings, NamedMpkFileRecorder, Recorder};
use burn_import::safetensors::SafetensorsFileRecorder;

type Backend = burn_ndarray::NdArray<f32>;

fn convert_model() {
    let device = Default::default();

    // Load from Safetensors
    let recorder = SafetensorsFileRecorder::<FullPrecisionSettings>::default();
    let record = recorder
        .load("./conv2d.safetensors".into(), &device)
        .expect("Should decode state successfully");

    // Save to Burn's binary format
    let recorder = NamedMpkFileRecorder::<FullPrecisionSettings>::default();
    recorder
        .record(record, "model.mpk".into())
        .expect("Failed to save model record");
}

// In your application code
fn load_model() -> Net<Backend> {
    let device = Default::default();

    // Load from Burn's binary format
    let record = NamedMpkFileRecorder::<FullPrecisionSettings>::default()
        .load("./model.mpk".into(), &device)
        .expect("Should decode state successfully");

    Net::<Backend>::init(&device).load_record(record)
}
}

Note: For examples of pre-converting models, see the examples/import-model-weights directory in the Burn repository.

Advanced Configuration Options

Framework-Specific Adapters

When importing Safetensors models, you can specify an adapter type to handle framework-specific tensor transformations. This is crucial when importing models from different ML frameworks, as tensor layouts and naming conventions can vary:

#![allow(unused)]
fn main() {
let device = Default::default();

// Create load arguments with framework-specific adapter
let load_args = LoadArgs::new("model.safetensors".into())
    .with_adapter_type(AdapterType::PyTorch); // Default adapter

// Load with the specified adapter
let record = SafetensorsFileRecorder::<FullPrecisionSettings>::default()
    .load(load_args, &device)
    .expect("Should decode state successfully");
}

Available Adapter Types

Troubleshooting and Advanced Features

Key Remapping for Different Model Architectures

If your Burn model structure doesn't match the parameter names in the Safetensors file, you can remap keys using regular expressions:

#![allow(unused)]
fn main() {
let device = Default::default();

// Create load arguments with key remapping
let load_args = LoadArgs::new("model.safetensors".into())
    // Remove "conv" prefix, e.g. "conv.conv1" -> "conv1"
    .with_key_remap("conv\\.(.*)", "$1");

let record = SafetensorsFileRecorder::<FullPrecisionSettings>::default()
    .load(load_args, &device)
    .expect("Should decode state successfully");

let model = Net::<Backend>::init(&device).load_record(record);
}

Debugging with Key Inspection

To help with troubleshooting import issues, you can enable debugging to print the original and remapped keys:

#![allow(unused)]
fn main() {
let device = Default::default();

// Enable debug printing of keys
let load_args = LoadArgs::new("model.safetensors".into())
    .with_key_remap("conv\\.(.*)", "$1")
    .with_debug_print();  // Print original and remapped keys

let record = SafetensorsFileRecorder::<FullPrecisionSettings>::default()
    .load(load_args, &device)
    .expect("Should decode state successfully");
}

Automatic Handling of Non-Contiguous Indices

The SafetensorsFileRecorder automatically handles non-contiguous indices in model layer names. For example, if the source model contains indices with gaps:

"model.layers.0.weight"
"model.layers.0.bias"
"model.layers.2.weight"  // Note the gap (no index 1)
"model.layers.2.bias"
"model.layers.4.weight"
"model.layers.4.bias"

The recorder will automatically reindex these to be contiguous while preserving their order:

"model.layers.0.weight"
"model.layers.0.bias"
"model.layers.1.weight"  // Reindexed from 2
"model.layers.1.bias"
"model.layers.2.weight"  // Reindexed from 4
"model.layers.2.bias"

Partial Model Loading

You can selectively load weights into a partial model, which is useful for:

• Loading only the encoder from an encoder-decoder architecture
• Fine-tuning specific layers while initializing others randomly
• Creating hybrid models combining parts from different sources

Support for Enum Modules

The SafetensorsFileRecorder supports models containing enum modules with new-type variants. The enum variant is automatically selected based on the enum variant type, allowing for flexible model architectures.