Running AI Workloads Across NVIDIA, AMD, and Multiple Clouds Without RefactoringA technical look at hardware-agnostic AI infrastructure — running training on NVIDIA H100s and inference on AMD MI300X across multiple clouds.6 min read·flexai.hashnode.devApr 30
Fine-Tuning a 32B Legal LLM That Outperformed a Frontier Model at 4× Lower Serving CostLegML fine-tuned a 32B legal LLM using GRPO and supervised fine-tuning that beat a frontier model on legal benchmarks.6 min read·flexai.hashnode.devApr 21
LLM Inference GPU Sizing: How to Choose the Right GPU for Your Model and TrafficHow to size GPUs for LLM inference in production. Walk through real latency and cost tradeoffs for LLaMA 3.1 8B across L40, H100, and H200 — with toke5 min read·flexai.hashnode.dev1Apr 14
Why AI Infrastructure Software Is Harder Than Hardware — Lessons from Building Aurora and FlexAIFlexAI CEO Brijesh Tripathi shares lessons from 25 years at NVIDIA, Apple, Tesla, and Intel — including delivering the Aurora supercomputer7 min read·flexai.hashnode.devApr 7
Heterogeneous AI Compute: Why Mixed NVIDIA, AMD, and TPU Clusters Are Harder Than They LookA practitioner's look at the runtime, scheduling, and deployment challenges of running AI training and inference across different architectures.8 min read·flexai.hashnode.devApr 3
