Ragweld Crucible

Operator range planner for estimating training cost, VRAM fit, and GPU availability across providers.

Crucible is the estimator workbench behind ragweld.com/crucible. It combines model metadata, hardware reference data, provider pricing, and planning heuristics so operators can compare likely training ranges before committing a run. Instead of guessing whether a fine-tune fits in VRAM or how much it will cost across providers, Crucible gives you a concrete range estimate upfront.

Key Features

• Training cost estimation across GPU providers and hardware configurations
• VRAM fit calculator with model metadata and hardware reference data
• Math code workbench for inspecting the implementation behind planner calculations
• Model catalog with automated refresh scripts for up-to-date metadata
• Shareable planning sessions with URL-encoded state for team collaboration

Technical Architecture

The front-end is a React + TypeScript + Vite application with estimator logic, normalization rules, and planner math in a dedicated engine module. It ships with reference datasets for model catalogs, GPU specs, and pricing fallback data.

This repository is a public mirror of the crucible subtree from the larger ragweld.com site repo, containing the full front-end app, estimator logic, data catalogs, and tests. The live production surface also depends on Netlify Functions from the parent repo for dynamic pricing data. Tests run via Vitest.