TildAlice – Tech Blog for Developers
Python, AI, Deep Learning, Data Analysis, Computer Vision, Reinforcement Learning โ tips and tutorials
Latest Posts
-
RLHF vs DPO: Training Cost Drops 68% in Real Migration
RLHF to DPO migration cut our 7B model training cost from $12.4K to $3.95K. Here's what broke, what worked, and the one dataset bug that tanked accuracy.
-
OpenCV to Albumentations: 3x Faster Augmentation Pipeline
Migrate from OpenCV to Albumentations for 3x faster image augmentation. Compare performance, GPU support, and integration tips for CV pipelines.
-
PyTorch vs TensorFlow 2026: Why Framework Wars Distract
Compare PyTorch vs TensorFlow in 2026 and discover why the framework debate misses the point. Build better models by focusing on what truly matters.
-
Real-Time FFT Pipeline: Vibration to Alert in 100 Lines
Build a streaming FFT pipeline in 100 lines: raw accelerometer data to bearing fault alerts in 2 seconds. No Kafka, just NumPy and signal processing.
-
PCA vs t-SNE vs UMAP: Real Performance on 10K Samples
Compare PCA, t-SNE, and UMAP on 10K samples with runtime benchmarks, visual quality analysis, and practical guidance for your data pipeline.
-
mypy vs pyright vs Pyre: 47% Error Detection Gap
I ran the same 60-bug test corpus through mypy, pyright, and Pyre. pyright caught 58 errors, mypy found 34, Pyre only 29. Real accuracy benchmarks.
-
Q-Learning from Scratch: 50-Line Agent Beats Random by 94%
Write a 50-line Q-Learning agent that beats random policy by 94% on FrozenLake. Hyperparameter mistakes, convergence curves, and why it fails on CartPole.
-
EAL6+ vs EAL5+: Why Hardware Wallet Chip Certification Matters
Common Criteria ratings aren't marketingโEAL6+ chips resist side-channel attacks that compromise EAL5+. Here's why OneKey's certification actually matters.