tinker

Streamline King Legacy on PC with a fast desktop script hub featuring 500+ Lua scripts and customizable toggles.

Automate RL fine-tuning by letting an AI agent run and evaluate multiple training experiments to find stable post-training configurations.

Browser-based PHP REPL for any Laravel Herd project. Write, save, and run snippets, see output rendered instantly.

Christ-centered LLM post-training research using Scripture-governed objectives, Tinker LoRA, and causal evaluation.

Find out which process is making your Windows mouse cursor spin — and why. Single-file Python GUI + CLI, no dependencies.

Panel-of-experts LoRA vs Qwen3 thinking on Qwen3-30B-A3B: accuracy and completion-length comparison, plus an exploratory RL follow-up

Powerful REPL for the Laravel framework.

Open-source evolutionary post-training lab powered by the Tinker API

Agent-native MCP control plane for Tinker post-training workflows: datasets, 34 Cookbook recipes, training, evaluation, checkpoints, and exports.

Easily tweak your PHP code

A desktop GUI tool that extracts every line of text from a batch of PDFs (given as URLs in an Excel file) and exports it to CSV, with per-file timeouts and incremental batch saving.

한국어 사전학습과 SFT를 위한 1.09B 파라미터 풀스크래치 LLM — RoPE, KV Cache, BF16·8-bit AdamW 최적화, 데이터셋 캐싱 및 학습 모니터링 지원.

Run the PHP code in the browser. - 在浏览器中运行 PHP 代码。

Fine-tune open-source models with Tinker from inside Pi — managed improve loops, data prep, evals, smoke tests, deploy snippets, and checkpoint chat.

An open-source research interface for interactive LLM tuning workflows with Tinker: dataset registration, recipe selection, run monitoring, checkpoint inspection, behavior testing, and Hugging Face deployment.

Democratizing Reinforcement Learning for LLMs

A modular python toolkit for automated SOC triage, phishing analysis, threat intelligence, and digital forensics.

Native reinforcement learning for Apple MLX, with a Tinker-compatible API.

Experiments in fine-tuning and evaluating language models that make sequential, tool-assisted decisions.

Multi-threaded YOLO detection/tracking + speed-based distance logging prototype with a Tkinter GUI.