Fine-Tuning
ML/dictionary/fine-tuning
Definition
Continuing to train an existing model on new data, so the new patterns get baked into the weights. Distinct from RAG (which only changes the prompt) and prompting (which changes nothing).
Posts that use this term
- Troubleshooting local LLMs and keeping up
The catalog of common local-LLM failures: OOM, slow tok/s, garbage output, instruction drift, RAG miss, tool-call hallucination. Plus where to follow the field as it moves.
- Fine-tuning a model locally
When fine-tuning is the right answer (rarely) and how to do it on consumer hardware: LoRA, QLoRA, MLX-LM, Unsloth. A worked example fine-tuning Llama 3.2 3B on a 16GB Mac.
- Local agents and tool use
Function calling on open models in 2026: which models actually work (Qwen 2.5, Hermes 3, Llama 14B+), why local agents fail when they fail, and how to build defensive scaffolding around them.
- Every machine can run a local LLM (here's what fits)
Per-tier guide: 8GB integrated graphics, 16GB MacBook Air, 8/12/16/24/32GB VRAM PCs, 24/32/64/128/192GB Macs. Specific models, specific tok/s, specific configs. Every tier runs something useful.
- Picking a local model by task
The 2026 open leaders by task: coding (Qwen 2.5 Coder, DeepSeek-Coder), chat (Llama, Qwen, Mistral), small-model renaissance (Phi-3, Gemma 2), structured output, multimodal, embeddings.
- The local-LLM vocabulary
Parameters, B, dense vs MoE, base vs instruct, tokens, context window, chat template, GGUF, quantization suffixes. After this post you can read any HuggingFace model card.
- The pitch for local LLMs in 2026
Why every engineer should run a local LLM in 2026: privacy, zero marginal cost, lower latency, no rate limits, and offline. Even a 16GB MacBook Air runs Llama 3.2 3B at 30 tok/s.
- What leaves your machine when you use AI
What providers actually see, log, and retain when you call an LLM API in 2026. What 'we don't train on your data' really means, free vs paid tier differences, and when local is the only safe option.
- The major LLMs in 2026
A tour of the closed frontier models (Claude, GPT, Gemini) and the open weights (Llama, Qwen, DeepSeek, Mistral). What 'B' means, what each is good at, and which size to actually run.
- Prompt, RAG, fine-tune: three ways to shape a model
Three levers for shaping what an LLM does: prompting (ask better), RAG (give it the right context), fine-tuning (change the weights). What each costs, what each fixes, and how to pick.
- RAG: giving a model memory it doesn't have
RAG is the pattern of fetching relevant text from a search system and putting it in the LLM's context window before asking your question. Not magic, not fine-tuning — just better prompts.
- From models to LLMs
An LLM is one kind of ML model — trained on text, predicts the next token. That single trick at scale gets you ChatGPT, and also explains where it breaks.
- How a model learns: training and inference
Training is the expensive one-time event where a model's numbers get tuned. Inference is the cheap repeated use afterwards. The gap in cost is enormous, and it shapes the whole industry.