TildAlice – Tech Blog for Developers
Python, AI, Deep Learning, Data Analysis, Computer Vision, Reinforcement Learning โ tips and tutorials
Latest Posts
-
Model Drift Detection Failed Silently: Evidently AI Fix
Fix Evidently AI's silent drift detection failures with schema validation. Real production errors, Wasserstein metrics, and monitoring setup.
-
Pre-commit Hooks That Break CI: 5 Fixes from 50+ Repos
Your pre-commit hook passes locally but CI fails with the same check. Here's why it happens and 5 fixes that actually work.
-
ONNX Runtime Mobile: 8ms Inference on iPhone 13
Cut mobile inference from 200ms to 8ms by switching to ONNX Runtime. Benchmarks, CoreML quirks, and when TFLite still wins.
-
GARCH vs LSTM for Bitcoin Volatility Forecasting
GARCH(1,1) beat LSTM by 32% on Bitcoin volatilityโwhy the 1986 model wins and when neural networks might actually outperform it.
-
Python __slots__: 40% Memory Savings and Interview Gotchas
Cut Python class memory by 40% with __slots__โplus the 3 gotchas interviewers use to catch candidates who only read the docs.
-
Triton vs TorchServe: How I Cut $800/Month in GPU Costs
Triton cut my inference costs 52% vs TorchServe. GPU utilization jumped from 23% to 81%. Here's what profiling actually revealed.
-
GQA Review: Grouped Query Attention for Faster LLM Inference
GQA reduces LLM inference memory 8x with only 0.1 ROUGE loss. How grouped KV heads and 5% uptraining work in Llama 2 and Mistral.
-
Why Backtesting Returns Lie: 5 Overfitting Traps
Your 47% backtest return probably measures noise. Five overfitting traps that turn algo strategies into expensive lessonsโwith fixes.