TL;DR Training deep learning models means solving an optimization problem: The model is incrementally adapted to minimize an objective function. The optimizers used for training deep learning models are based on gradient descent, trying to shift the model’s weights towards the objective function’s minimum. A range of optimization algorithms is used to train deep learning models, each aiming to address a particular shortcoming of the basic gradient descent approach. Stochastic Gradient Descent (SGD) and Mini-batch Gradient Descent speed up training and are suitable for larger da ..read more

Here’s a quarterly product newsletter to keep you up-to-date with all the changes in Neptune. Check what happened in the last 3 months.   New ? 1. Scatter plots If you have two metrics or parameters that you wish to compare or see how they relate to each other throughout the runs, you can now create a scatter plot.  See an example of a scatter plot in the Neptune app. 2. Distraction-free mode  You can now view run dashboards and compare views in distraction-free mode.  3. Integrations & supported tools We have 4 new integrations: Airflow  Modelbit (comm ..read more

We are excited to announce that Neptune and Modelbit have partnered to release an integration to enable better ML model deployment and experiment tracking. Data scientists and machine learning engineers can use the integration to train and deploy machine learning models in Modelbit while logging and visualizing training progress in Neptune. If you are not already familiar, Neptune is a lightweight experiment tracker for MLOps. It offers a single place to track, compare, store, and collaborate on experiments and models.  Neptune’s UI Modelbit is a machine learning platform that makes deplo ..read more

I always look forward to sharing these updates with you! Hope you’ll find something here that can enhance your workflow. Here’s what we’ve released in the last quarter. New ? 1. MosaicML Composer integration New integration alert! With the Neptune-Composer integration, you can automatically log your Composer training metadata to Neptune. See the example project here. 2. Support for Seaborn figures We’re expanding the list of supported visualizations. Now, you can log and display Seaborn figures as PNG images. 3. Documentation updates We documented the Neptune Quer ..read more

Before you dive into 2024, have a look at what we released in Neptune in the last 3 months. New ? 1. MLflow plugin The new Neptune-MLflow integration allows you to send your metadata to Neptune while using the MLflow logging code. It should be especially handy when you already use MLflow in some of your projects but you’d like to enhance the tracking with Neptune’s functionality.  2. Documentation updates To make your life easier, we’re constantly improving our documentation.  There’s a new API index available. This page lists all functions, parameters, and constants exposed by ..read more

TL;DR GPUs can greatly accelerate deep learning model training, as they are specialized for performing the tensor operations at the heart of neural networks. Since GPUs are expensive resources, utilizing them to their fullest degree is paramount. Metrics like GPU usage, memory utilization, and power consumption provide insight into resource utilization and potential for improvement. Strategies for improving GPU usage include mixed-precision training, optimizing data transfer and processing, and appropriately dividing workloads between CPU and GPU. GPU and CPU metrics can be moni ..read more

TL;DR Chatbots based on LLMs can solve tasks they were not trained to solve either out-of-the-box (zero-shot prompting) or when prompted with a couple of input-output pairs demonstrating how to solve the task (few-shot prompting). Zero-shot prompting is well-suited for simple tasks, exploratory queries, or tasks that only require general knowledge. It doesn’t work well for complex tasks that require context or when a very specific output form is needed. Few-shot prompting is useful when we need the model to “learn” a new concept or when a precise output form is required. It’s also a ..read more

TL;DR LLMOps involves managing the entire lifecycle of Large Language Models (LLMs), including data and prompt management, model fine-tuning and evaluation, pipeline orchestration, and LLM deployment. While there are many similarities with MLOps, LLMOps is unique because it requires specialized handling of natural-language data, prompt-response management, and complex ethical considerations. Retrieval Augmented Generation (RAG) enables LLMs to extract and synthesize information like an advanced search engine. However, transforming raw LLMs into production-ready applications presents ..read more

TL;DR MLflow is a popular experiment-tracking and end-to-end ML platform Since MLflow is open source, it’s free to download, and hosting an instance does not incur license fees Hosting MLflow requires multiple infrastructure components and comes with maintenance responsibilities, the cost of which can be difficult to estimate On AWS, which offers various options for hosting MLflow, a medium-sized instance comes in at about $200 per month, plus storage and data transfer costs MLflow is well-regarded as an experiment-tracking platform. Since it’s open source, you can download i ..read more

TL;DR Deep learning models exhibit excellent performance but require high computational resources. Optimization techniques like pruning, quantization, and knowledge distillation are vital for improving computational efficiency: Pruning reduces model size by removing less important neurons, involving identification, elimination, and optional fine-tuning. Quantization decreases memory usage and computation time by using lower numeric precision for model weights. Knowledge distillation transfers insights from a complex “teacher” model to a simpler “student” model, maintaining performa ..read more