Machine Learning
Mistral-7B-v0.3 QLoRA on Modal A100-40GB: nf4 + bf16_compute Beat My RunPod H100 Spot Cost Per Step
TL;DR: For a Mistral-7B-v0.3 QLoRA fine-tune at sequence length 2048 and micro-batch 4, a Modal A100-40GB container running bitsandbytes nf4 with bfloat16.
torch.compile in PyTorch 2.5: Where the Speedup Comes From and Where It Disappears
PyTorch 2.5 made torch.compile good enough that you can drop it into a real training script and expect a speedup most of the time.
How to Automate Hyperparameter Tuning in PyTorch With Optuna
I still remember the early days of my machine learning career, sitting in front of a terminal at 2 AM, manually tweaking learning rates, batch sizes, and.
How to Convert PyTorch Models to ONNX Format for Faster Inference
I remember the first time I deployed a PyTorch model to production. I wrapped a beautifully trained ResNet model in a Flask API, spun up a Docker.
The Stable Kaggle CLI Fixes My Biggest Authentication Headache
I was staring at my terminal at 11pm last Tuesday, watching a GitHub Actions runner fail for the third time. The error was always the same.
Massive AI Models Are Failing. Small Fast.ai Builds Win.
I was staring at my AWS bill last Tuesday, trying to figure out how a simple image classification microservice managed to rack up $840 in three weeks.
Meta’s $100B AMD Pact Actually Fixes PyTorch’s Biggest Headache
The Monopoly Tax is Getting Old I spent three hours yesterday trying to provision a single H100 instance on AWS. Three hours. For one node.
TensorRT Just Fixed Local Image Generation
Running modern, heavy diffusion models locally has felt like trying to stuff a mattress into a compact car for months now. You Learn about TensorRT News.
Hiding Android Malware in Hugging Face Repos
I spent my entire Tuesday morning cleaning up a mess because a junior developer treated Hugging Face like a trusted package manager. It isn’t.
Ditching Heavy Transformers for Static Embeddings
Well, I have to admit, I actually stumbled upon this solution by accident. There I was, staring at our AWS bill at 2am last Tuesday, trying to figure out.