Error: Prediction failed: CUDA out of memory. Tried to allocate 5.71 GiB (GPU 0; 14.61 GiB total capacity; 8.08 GiB already allocated; 2.04 GiB free; 11.60 GiB reserved in total by PyTorch) If reserved memory is >>

[SangVoM]

I am using this library to scale images but when I use this code I get error:

Error: Prediction failed: CUDA out of memory. Tried to allocate 5.71 GiB (GPU 0; 14.61 GiB total capacity; 8.08 GiB already allocated; 2.04 GiB free; 11.60 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

this is my code :

module.exports = async function replicateImage(imagePath) {
    function imageToBase64(imagePath) {
        try {
            const imageBuffer = fs.readFileSync(imagePath);
            const base64 = imageBuffer.toString("base64");
            const mimeType = "image/png";
            return `data:${mimeType};base64,${base64}`;
        } catch (error) {
            console.error('Error converting image to base64:', error);
            return null;
        }
    }
    async function scaleImage(imagePath) {
        const replicate = new Replicate({
            auth: process.env.REPLICATE_API_TOKEN
        });
        const model = 'nightmareai/real-esrgan:42fed1c4974146d4d2414e2be2c5277c7fcf05fcc3a73abf41610695738c1d7b';
        const input = {
            image: imageToBase64(imagePath),
            scale: 8
        };
        try {
            console.log('Start scaling image...');
            const result = await replicate.run(model, { input });
            const outputImagePath = path.join(path.dirname(imagePath), `scaled-${path.basename(imagePath)}.png`);
            const response = await axios.get(result, { responseType: 'arraybuffer' });
            fs.writeFileSync(outputImagePath, response.data);
            console.log('Done scaling image.')
            return outputImagePath;
        } catch (error) {
            console.error('Error scaling image:', error);
        }
    }
    try {
        return await scaleImage(imagePath);
    } catch (error) {
        throw error;
    }
}

Can you help me ?

[mattt]

Hi @SangVoM, the error describes a problem in the model itself. It looks like scaling an image by 8x exceeds the memory capacity of the GPU used to run the model, so you might consider scaling by a smaller factor, like 4x.

[SangVoM]

Okay @mattt