Giorgio Gilestro
8da0dac007
Adds the union-preserving recombination operator that llm_merge lacked (E8's max,
not mean): keep each specialist LoRA intact and SELECT the right one per prompt
(MoE router: oracle, or training-free nearest-centroid over base embeddings) or
per module (max_merge = winner-take-all by delta norm). src/llm/moe.py, kind
llm_moe, reuses the cached specialists.
Result — a clean regime boundary for "merge, don't average":
- 0.5B: union wins. Routing 0.74 / worst-family 0.43 > soup 0.64 / 0.26, with no
dilution (recovers each specialist's own-family peak). E8's max > mean in real
weights, because at a weak base averaging dilutes.
- 7B (Imperial CX3, L40S, 9 min): the ordering INVERTS. Fusion wins — soup 0.87 >
routing 0.84 > max_merge 0.78. Routing is capped at the best parent per family;
fusion blends and, given a capable base, COMPOSES beyond any parent (soup lists
0.62 > spec 0.57). Selection can't synthesise better than its best component;
averaging-that-composes can.
So "merge, don't average" (E4/E8) is a weak-parent / small-model law, not
universal: union wins under dilution, fusion wins under composition. Refines E8
(its additive-landscape max>mean assumed no compositional headroom). The operator
to want is fusion-that-composes + offspring selection = the directed-sex ideal
(E10) — the natural next experiment.
Honest riders: the learned router is trivially perfect (lexically-distinct
families), and router-free max_merge is the weakest union (not input-adaptive).
+2 router unit tests (127 green). Results in results/llm_moe{,_hpc}/ (parquet
gitignored per the reproducibility contract).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
44 lines
2.2 KiB
Makefile
44 lines
2.2 KiB
Makefile
# Layer 1 + Layer 1.5 automation. The uv venv (built from the committed uv.lock) is the
|
|
# reproducibility source of truth; every target runs inside it via `uv run`.
|
|
|
|
.PHONY: env env-neural env-mnist env-llm test layer1 layer2 neural mnist llm figures clean
|
|
|
|
env: ## build .venv from the committed lockfile
|
|
uv sync --extra dev
|
|
|
|
env-neural: ## add the Layer 1.5 torch stack (GPU; Stage C onward)
|
|
uv sync --extra dev --extra neural
|
|
|
|
env-mnist: ## add torchvision for the real-MNIST confirmation tier
|
|
uv sync --extra dev --extra neural --extra mnist
|
|
|
|
test: ## correctness tests + scientific-validation tests (the spine of trust)
|
|
uv run pytest
|
|
|
|
layer1: ## run experiments E1-E6 + the learning-kernel bridge (analytic)
|
|
for e in E1 E2 E3 E4 E5 E6 E7 E8 E9 E10 E11 kernel_sharpen kernel_smooth; do uv run python -m knowledge.experiment configs/layer1/$$e.yaml; done
|
|
|
|
neural: ## run Layer 1.5 synthetic neural experiments (excludes the heavy MNIST tier)
|
|
for c in configs/neural/*.yaml; do case "$$c" in *mnist*) ;; \
|
|
*) uv run python -m neural.experiment "$$c" ;; esac; done
|
|
|
|
mnist: ## run the real-MNIST confirmation tier (needs env-mnist; downloads MNIST)
|
|
uv run python -m neural.experiment configs/neural/mnist_collapse.yaml
|
|
|
|
env-llm: ## add the LLM stack for the Layer-2 prototype (GPU; transformers/peft)
|
|
uv sync --extra dev --extra neural --extra llm
|
|
|
|
llm: ## run the LLM prototypes: merge (fusion) + moe (union) recombination
|
|
uv run python -m llm.experiment configs/llm/merge.yaml
|
|
uv run python -m llm.experiment configs/llm/moe.yaml
|
|
|
|
layer2: neural ## alias: Layer 1.5 is the current Layer-2 deliverable (LLM rung deferred)
|
|
|
|
figures: ## regenerate figures from committed results
|
|
for e in E1 E2 E3 E4 E5 E6; do MPLBACKEND=Agg uv run python figures/plot_$$e.py; done
|
|
for p in figures/plot_*.py; do case "$$p" in */plot_E[1-6].py|*/_*) ;; \
|
|
*) [ -e "$$p" ] && MPLBACKEND=Agg uv run python "$$p" ;; esac; done
|
|
|
|
clean: ## remove caches and generated results (keeps committed manifests)
|
|
rm -rf .pytest_cache **/__pycache__
|
|
find results -type f ! -name '.gitkeep' -delete 2>/dev/null || true
|