NVidia’s Llama 3.1 Nemotron 70B is outperforming larger and closed models on key reasoning tasks. In this post, Brendan tests it against a long-unsolved challenge: consistent, in-character roleplay with zero internal monologue or user coercion—and finds it finally up to the task.

Large language models can write poetry and solve logic puzzles—but fail at tasks like counting letters or doing math. Here’s why, and what it tells us about their design.

Use Ollama to compare multiple LLMs side-by-side on a single GPU pod—perfect for fast, realistic model evaluation with shared prompts.

Discover how to boost your LLM inference performance and customize responses using SGLang, an innovative framework for structured LLM workflows.

Learn how to optimize your serverless GPU deployment on Runpod to balance latency, performance, and cost. From active and flex workers to Flashboot and scaling strategy, this guide helps you build an efficient AI backend that won’t break the bank.

Runpod Serverless now supports multi-GPU workers, enabling full-precision deployment of large models like Llama-3 70B. With optimized VLLM support, flashboot, and network volumes, it's never been easier to run massive LLMs at scale.

Runpod introduces Serverless CPU: high-performance VM containers with customizable CPU options, ideal for cost-effective and versatile workloads not requiring GPUs.