Fonderie · Atelier 01

Affinage refining

Training and fine-tuning open models on domain data, packaged as customised local models.

Affinage is the refining of molten metal: removing impurities until it reaches a specified composition.

État Draft. The run record, hyperspectral data handling, conversion of the HyperSIGMA base encoders to safetensors, three customisation methods and export with a model card work and are tested on synthetic data. No model trained on real data has been produced or released. This site is still an outline; detailed sections stay marked coming soon until they are written.

The outline GitHub ↗

Fonderie mark: a crucible pouring into an ingot mould whose cells read as a weight matrix

SommaireWhat this site will hold

PérimètreIntended scope

What Affinage is meant to do. The scope is written before the code, so the code can be held to it; none of it is implemented yet.

  • Configuration-driven training and fine-tuning runs of open-weight models, full and parameter-efficient (for example LoRA), from a base model pinned to a commit hash.
  • Data preparation with a manifest and checksums, and a recorded split that Essai can check for train–test overlap.
  • A run record written next to every output.
  • Export of the result as a customised local model: merged safetensors weights, an adapter, or a quantised file for a local runtime, each with a model card.

Not in scope

Hosted training services; weight inspection and surgery (Burin); benchmark evaluation (Essai).

JalonsFirst milestones

  1. Run-record schema and writer, shared in form with Essai.

    Coming soon
  2. A fine-tuning run on a tiny model and synthetic data that completes on CPU inside the test suite.

    Coming soon
  3. Export with a generated model card.

    Coming soon

The journal →

La chaîneOne of three ateliers

  1. Affinage refining

    Training and fine-tuning open models on domain data, packaged as customised local models.

    This site

    Status: draft
  2. Burin engraving

    Inspecting and editing the weights of open models.

    Status: scaffold
  3. Essai assay

    Reproducible evaluation of models on scientific tasks.

    Status: scaffold

Every atelier follows the same rules for records, artefacts and safe loading: the Fonderie method ↗