---
title: "PostTrainLLM docs"
description: "Mac-local LLM specialist factory — training, inference, evals, systems notes, and learning paths."
---

# PostTrainLLM documentation

PostTrainLLM is a **Mac-local learning lab and specialist factory**. The
current owner learning route is the [13-week inference-systems
sprint](learn/inference-systems-13w.md), beginning 28 September 2026. The
retained factory loop is target → data → post-training → eval → package →
report. The docs keep those two jobs separate so an old plan never looks like
today's assignment.

## Start learning

1. [Open the course workspace](learn.md) for the 13-week route and the
   prerequisite path.
2. [Open the single Next Session](learning-progress.md#next-session) for the
   current exercise, evidence to save, and recall schedule.
3. Use the [ground-up curriculum](learn/curriculum.md) only when an entry
   check reveals a gap. The [full learning index](learn/README.md) remains
   available for deeper reading.

The first week asks you to predict tensor shapes and causal influence, trace
token IDs to logits, and run one bounded CPU reference check. No owner
checkpoint or completed mastery gate is implied by this navigation.

## Find a document

| Shelf | Use it for |
|---|---|
| [Learning library](learn/README.md) | Foundations, systems concepts, case studies, and practical paths |
| [Factory handbook](factory/README.md) | Run schema, eval protocol, packaging, and reports |
| [Techniques and recipes](techniques/README.md) | Retained methods and bounded recipes |
| [Attempt ledger](attempt-ledger.md) | What worked, failed, regressed, or received a final disposition |
| [Document library map](library.md) | The full reference, evidence, research, PRD, parked, and archive shelves |

The [project status](https://github.com/PostTrainLLM/posttrainllm/blob/main/PROJECT_STATUS.md)
and [closure rule](NEXT.md) govern any new experiment. Historical TODOs and
PRDs are context, not active tasks. For maintainer reading order, see the
[golden path](README.md); for conflicting documents, use the
[status registry](doc-status.md).

Every published page also has a Markdown mirror: append `.md` to its docs URL.