Hubs
The guides and tools for one recurring problem, bundled onto a single page instead of spread across the blog and tools archive.
GPU / CUDA Out of Memory & Device Errors
Your training run dies with a CUDA OOM, a device-side assert, or a tensor stuck on the wrong device. These are the specific PyTorch and Hugging Face errors that cause it, in the order they usually show up, plus a calculator to estimate VRAM before you hit the wall again.
5 articles · 1 tools
Hugging Face Model Loading & Caching Errors
Separate from the GPU memory wall: these are the errors that hit before training even starts, disk space filling up in the model cache, a tokenizer with no pad_token, or an import that moved between huggingface_hub versions.
3 articles · 0 tools
scikit-learn Errors
The scikit-learn errors that show up once real, messy data hits a pipeline: NaNs that break a fit, a string column an estimator can’t convert, a model used before it’s trained, or a version mismatch after an upgrade.
6 articles · 1 tools
pandas Data-Wrangling Errors
Mostly pandas 2.0-era breakage: methods like .append() and .iteritems() that got removed, datetime columns that refuse to parse, and the SettingWithCopyWarning that shows up the moment you chain a filter and an assignment.
4 articles · 0 tools
Understanding Neural Networks From Scratch
Before debugging a framework, understanding what it’s actually computing: what supervised/unsupervised/RL mean, a forward pass and backprop implemented in plain NumPy, and calculators for the matrix math and parameter counts underneath it.
2 articles · 2 tools
LLM Pipelines, Embeddings & Retrieval
Building something on top of embeddings, retrieval, or multiple LLM agents: real applied pipelines over medical literature and campaign data, the vector-search (FAISS) errors that show up once an index grows past a toy size, and the LangChain/LlamaIndex import errors that block getting started at all.
6 articles · 3 tools
Python ML Environment & Import Errors
The generic errors that stop an ML script from even running, independent of which model or framework it uses: a binary-incompatible numpy build, XGBoost rejecting a DataFrame’s dtypes, an unpacked None from a function that silently failed, or OpenCV missing entirely.
4 articles · 0 tools