ollama-model-tuner
от gblockchainnetwork
DevOps & CloudКратко
Optimize Ollama models/prompts using local datasets, eval metrics, and iterative tuning.
Описание
Что делает этот навык
Tunes Ollama models locally using custom datasets, prompt engineering, and LoRA fine-tuning. Supports A/B testing prompts, customizing Modelfiles, and benchmarking model performance against eval metrics. No cloud services or external APIs required.
Runs the entire tuning loop on local hardware, so sensitive training data never leaves the machine and there are no per-token API costs.
Когда его использовать
- Fine-tuning a local llama3 model for document classification
- A/B testing system prompts to find the best performer for a chatbot
- Adding a LoRA adapter trained on company-specific terminology
- Benchmarking response quality before and after Modelfile changes
- Iterating on prompt templates using a local JSONL eval dataset
Пример сценария
User asks: tune llama3 on my support ticket dataset for intent classification
- 1
Load JSONL dataset from ./data.json
- 2
Run scripts/tune.py with --model llama3 and --task classification
- 3
Apply prompt templates from prompts/system.md
- 4
Execute eval loop and record accuracy metrics
- 5
Output tuned Modelfile and benchmark report
Fine-tuned Modelfile with updated system prompt and eval metrics summary
Требования
- Ollama 0.3+ installed locally
Установка
clawhub install gblockchainnetwork/ollama-model-tunerВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.