Skip to content
PostTrainLLM docs
Esc
navigateopen⌘Jpreview
On this page

Doc map — where things moved

This restructure split, merged, and archived a few docs. Use this table to find where the content of any old path now lives.

If you came here looking for a doc that used to exist, find the old path in the left column → click the new path.

Split

The 1,400-line master roadmap was broken into a folder. Every section of the original lives at exactly one of the new paths:

Old path New path
docs/single_machine_roadmap.md (Part 1, Tier 1) docs/roadmap/tier1.md
docs/single_machine_roadmap.md (Part 1, Tier 2) docs/roadmap/tier2.md
docs/single_machine_roadmap.md (Part 1, Tier 3) docs/roadmap/tier3.md
docs/single_machine_roadmap.md (Part 1, Tier 4) docs/roadmap/tier4_skip.md
docs/single_machine_roadmap.md (Part 2, categories) docs/roadmap/categories.md
docs/single_machine_roadmap.md (Part 3, top-10 order) docs/roadmap/recommended_order.md
docs/single_machine_roadmap.md (Part 4, datasets) docs/roadmap/datasets.md
docs/single_machine_roadmap.md (Part 5, recent research) absorbed into docs/PLAN.md §4; archived at docs/archive/recent_research.md
docs/single_machine_roadmap.md (Part 6, phased plan) docs/roadmap/phased_plan.md
docs/single_machine_roadmap.md (Part 7, blockers) docs/roadmap/blockers.md
docs/single_machine_roadmap.md (honest summary) docs/roadmap/honest_summary.md
docs/single_machine_roadmap.md (index / TOC) docs/roadmap/index.md

The training-pipeline doc was split by phase:

Old path New path
docs/training_phases.md (Phase 1: Pretrain) docs/training/pretrain.md
docs/training_phases.md (Phase 2: SFT) docs/training/sft.md
docs/training_phases.md (Phase 3: DPO) docs/training/dpo.md
docs/training_phases.md (end-to-end + reading) docs/training/index.md

Merged (lossless)

Content of these docs now lives as appendices of the canonical home; the original moved to docs/archive/.

Old path New home Archived at
docs/evaluation.md docs/audits/validation_report.md (appendix) docs/archive/evaluation.md
docs/watch_the_model_think.md docs/techniques/interpretability.md (appendix) docs/archive/watch_the_model_think.md
docs/phase_9_10_status.md docs/roadmap/blockers.md (appendix) docs/archive/phase_9_10_status.md

Archived

Moved as-is to docs/archive/:

Old path New path
docs/annotated_transcript.md docs/archive/annotated_transcript.md
docs/parked_multi_model.md docs/archive/parked_multi_model.md
docs/shared_vs_native.md docs/archive/shared_vs_native.md

Root-level status and narrative docs, moved out of the repo root. None had been meaningfully updated since June; each is a historical record rather than current state, so they now sit with the other archived material:

Old path New path
HANDOFF.md docs/archive/HANDOFF.md
WHILE_YOU_SLEPT.md docs/archive/WHILE_YOU_SLEPT.md
MILESTONES.md docs/archive/MILESTONES.md
BLOG.md docs/archive/BLOG.md
NIGHTLY.md docs/training/nightly.md — a runner contract, not a status doc
STATUS.md deleted; it was a pointer stub. Current state is PROJECT_STATUS.md

Grouped (flat top level → topic folders)

The docs/ top level had grown to 94 loose files. Files that group unambiguously moved into topic folders; anything genuinely cross-cutting stayed at the top level. Old web URLs keep working — see below.

Moved into What went there
docs/performance/ FA2 notes, online softmax, KV-cache, memory tradeoffs, CPU/cold-start/checkpointing/YOCO results, perf audits and research, benchmark harness design, determinism contract
docs/techniques/ distillation, MoE, MTP, evolution strategies, pruning, quantization, precision, PEFT, LoRA guide, optimizers, GaLore, constrained generation, interpretability, speculative heads, StreamingLLM+KIVI
docs/audits/ audit_2026, docs-quality, exactness-completion, feature audit, history-coverage, test-coverage, validation report
docs/sessions/ dated session notes, Q&A log, Pace handoff, planner lock, v11 baselines, retrospective, drilldown, WWDC impact, first specialist findings
docs/integrations/ HuggingFace datasets, GitHub data, lm-eval-harness, Continue.dev provider, deploy
docs/guides/ model guide, training guide, study guide

docs/learning/ (one file) merged into docs/learn/.

URL redirects

Old web URLs (/docs/<old_slug>/) keep working via static redirects in browser/astro.config.mjs. So a link to /docs/single_machine_roadmap/ on social media still resolves; it just takes one extra hop to land at /docs/roadmap/.

Canonical homes (for DRY)

To avoid duplicating mechanics across docs, these are the canonical homes for repeated concepts. If you find an explanation of one of these outside its home doc, it should be a 1-2 sentence pointer + link, not a full explanation.

Concept Canonical home
bf16 / gradient accumulation / gradient checkpointing docs/performance/memory_tradeoffs.md
LoRA mechanics docs/techniques/lora_guide.md
MoE (mixture of experts) docs/techniques/moe.md
Distillation docs/techniques/distillation.md
MTP (multi-token prediction) docs/techniques/mtp.md
ES (evolution strategies) docs/techniques/evolution_strategies.md
Quantization (precision study) docs/techniques/precision.md
Quantization (Phase 9 status appendix) docs/roadmap/blockers.md
Interpretability (logit lens, attention vis, ablation) docs/techniques/interpretability.md
Training pipeline (pretrain → SFT → DPO) docs/training/
Post-training factory positioning docs/factory/post-training-factory.md
Docs golden path docs/README.md
CLI discovery and lab usage docs/cli-reference.md
Docs quality audit docs/audits/docs-quality-audit.md
Active/reference/archive status labels docs/doc-status.md
Attempt history / worked vs failed docs/attempt-ledger.md
External products reviewed / steals docs/external-products-reviewed.md
Method vs recipe / technique cards docs/techniques/
Owner learning pipeline docs/learning-pipeline.md
Owner learning progress docs/learning-progress.md
Single-machine roadmap + research docs/roadmap/
Open-source datasets (pretrain/SFT/DPO/code/math/eval) docs/roadmap/datasets.md

Was this page helpful?