ML Journey

A practical guide to HuggingFace fast tokenizers for ML engineers: how Rust-backed fast tokenizers differ from slow Python tokenizers, using offset mappings for NER and QA span alignment, high-throughput batched tokenisation with datasets.map and multiprocessing, sliding window tokenisation for long documents with stride and overflow, training a custom BPE vocabulary with the tokenizers library, and debugging gotchas around special tokens and sequence pair handling.

A guide to connecting Ollama with Obsidian for fully local AI-assisted note-taking: the Ollama community plugin for per-note summarisation and action item extraction, Smart Connections plugin for semantic indexing of the entire vault with nomic-embed-text and vault-wide RAG chat, Text Generator plugin via the OpenAI-compatible endpoint, a practical meeting notes workflow, building a queryable personal knowledge base, hardware recommendations, and getting started with just two ollama pulls.

A practical comparison of IA3 and LoRA for ML engineers: how IA3 activation scaling works versus LoRA weight updates, when each method wins (data volume, task type, adapter size), implementing IA3 with HuggingFace PEFT for classification and causal LM tasks, combining IA3 with 4-bit quantisation on consumer GPUs, and a decision framework for choosing between PEFT methods in production fine-tuning projects.

A complete guide to the official Ollama npm package in Node.js: installing with npm/yarn/bun, generate and chat with stream:false and stream:true, multi-turn CLI chatbot with readline, generating embeddings and computing cosine similarity, model management including pull with progress, delete, and ps, connecting to a remote Ollama server with a custom client, structured output using Zod schema passed directly to the format parameter, and image input for vision models.

A practical guide to sequence packing for ML engineers training LLMs: measuring padding waste and estimating speedup, greedy packing implementation with EOS separation, the attention leakage problem in naive packing, document-aware attention masks with Flash Attention cu_seqlens, TRL SFTTrainer packing configuration, and how to verify packing efficiency and model quality after implementation.

A complete Ollama REST API reference with curl examples for every endpoint: health check, /api/generate with streaming and options, /api/chat with multi-turn history and structured output format parameter, /api/embeddings, /api/tags to list models, /api/pull with progress streaming, /api/delete, /api/copy, /api/create from a Modelfile string, /api/ps for loaded models, /api/show for model details, and the OpenAI-compatible /v1/chat/completions, /v1/models, and /v1/embeddings endpoints.

A practical guide to multi-task learning for ML engineers: hard parameter sharing with task-specific heads, soft parameter sharing with cross-encoder regularisation, gradient cosine similarity for detecting negative transfer, homoscedastic uncertainty loss weighting, task sampling strategies, and an honest assessment of when multi-task training beats separate single-task baselines and when it does not.

A complete guide to Tabby, the open-source self-hosted coding assistant: what it does and how it compares to GitHub Copilot, installing via brew, binary, or Docker, running with built-in code models on CPU and GPU, connecting to VS Code and JetBrains IDEs with API token setup, model selection by hardware tier from 1.3B CPU to 13B GPU, enabling repository context indexing for project-aware completions, and running as a systemd service for persistent availability.

A practical guide to EleutherAI lm-evaluation-harness for ML engineers: CLI and Python API usage, running MMLU, HellaSwag, ARC and TruthfulQA, evaluating fine-tuned checkpoints, writing custom YAML tasks for domain benchmarks, understanding acc vs acc_norm vs mc2 metrics, and avoiding the prompt format mismatches and contamination issues that produce misleading benchmark numbers.

A practical guide to Mistral Nemo 12B: its distinctive technical features including 128K native context, the Tekken tokeniser, and strong multilingual training across 11 languages, hardware requirements at Q4_K_M (~7GB), when the 12B quality jump over 7–8B models is worth the extra VRAM, a Modelfile for long-context use, multilingual Python examples, and a clear comparison against Llama 3.2 8B and where Nemo wins.