28a2d8cf82
feat(edit): GPT-2 Conv1D edits, covariance ROME, atomic governor, Mixtral LongLoRA (v0.71.16)
...
Knowledge edit depth — 4-issue patch (validated on real gpt2 + SmolLM2-135M, RTX 3050).
- #251 GPT-2 (transformer.h / mlp.c_proj Conv1D) support in ROME/MEMIT/AlphaEdit:
transpose-aware _rank1_update + _alphaedit_project + MEMIT dim-check + PEFT unwrap.
- #250 covariance-preconditioned ROME via --cov-corpus (estimate_key_covariance +
C^-1 k* solve; fail-loud reject for non-rome; cwd-contained O_NOFOLLOW loader).
- #252 atomic EditGovernor increment: save_state merges this run's delta under the
cross-process lock (baseline-delta, mirrors namespace_pin).
- #147 Mixtral in the LongLoRA allowlist (is_mixtral_model + MixtralAttention
forward override + _SEPARATE_QKV_FAMILIES).
Tests: 13511 -> 13595 (+81 in tests/test_v07116.py). Full suite green; ruff clean.
2026-06-07 16:11:56 +05:00
d5f98c67d5
fix(loop): config-render, cmaes base-reuse, cost gate, energy hand-off, rank guard (v0.71.15)
...
#261 iterative_dpo._default_train_fn rendered output as a {dir: ...} mapping
that SoupConfig rejected; render it flat so the spawned soup train succeeds.
#246 CMA-ES merge now loads the base model once and reuses it across the
candidate population (_CachedBaseScorer) instead of reloading per candidate.
#245 soup loop estimate_cost wires run_cost.estimate_run_cost_usd off the last
completed run instead of a 0.0 placeholder; never crashes the daemon.
#244 soup train --track-energy --energy-out persists the measurement JSON so
soup bom emit --energy can attach it to an ML-BOM.
#170 --diagnose-gate is RANK-aware: gate once per cluster (RANK==0), not per node.
Tests: 13476 -> 13511 (+35 in tests/test_v07115.py). Validated end-to-end on
SmolLM2-135M / RTX 3050.
2026-06-07 13:58:18 +05:00
6d2b6399ab
docs: document soup bom emit --energy + credit @gittihub-jpg for #256
2026-06-06 19:55:46 +05:00
a2287dd6a5
feat(rag): RAFT span-mask trainer + RA-DIT auto-link + live steering + eval citation (v0.71.10)
...
Lifts the v0.62.0 RAG-family schema-only stubs to live, validated on SmolLM2-135M:
- #199 RAFT: data.format=raft trains answer-only (prompt span masked to -100,
[doc-N] citation ids, deterministic doc shuffle by raft_shuffle_seed); rows
whose prompt fills max_length are dropped with a warning. New utils/raft.py +
trainer/raft.py (RaftDataCollator + weighted-CE _RaftTrainer).
- #200 soup ra-dit: one-shot two-stage orchestrator (train retriever -> record
it as the generator's paired retriever -> train generator); a generator-stage
`soup train` with no retriever set auto-links the latest RA-DIT retriever from
the Registry. New utils/ra_dit_run.py + commands/ra_dit.py.
- #201 soup steer train/apply + soup serve --steer: live CAA/ITI/RepE fit from
{positive, negative} pairs + decode-time forward hook (transformers backend).
Lifts the steering.py apply_steering/build_steering_vector stubs.
- #202 soup eval citation + citation-span per-token loss boost + 7th `citation`
failure mode in soup diagnose. New commands/_eval_v07110.py +
diagnose/citation.py.
Review fixes (3 agents, all CRITICAL->LOW): markup-escaped autolink advisory;
shared enforce_under_cwd_and_no_symlink + O_NOFOLLOW on every new file read;
steering-artifact containment; honest RA-DIT docs (records pairing, no weight
fusion); public validate_ra_dit_config_path + render_raft_prompt; repe/iti
require >=2 pairs; eval citation --shuffle-seed.
Full suite: 13034 passed, 106 skipped (13142 collected). ruff clean.
2026-06-03 19:28:45 +05:00
96c339f184
feat(edit): live ROME/MEMIT/AlphaEdit + GRACE + NPO/SimNPO/RMU unlearn (v0.71.9)
...
Closes #193 , #194 , #196 , #197 , #203 .
- #194 utils/edit_kernels.py: covariance-free rank-1 ROME/MEMIT/AlphaEdit;
apply_edit live (load -> optimise residual -> rank-1 update -> save);
edit diff live before/after generation.
- #196 EditGovernorStore SQLite persistence + cross-process lock.
- #197 apply_edit consults the governor (check_can_edit before, record after).
- #203 GraceCodebook + apply_grace_edit + install_grace_hook + Registry kinds.
- #193 utils/unlearn_kernels.py (NPO/SimNPO/RMU) + live UnlearnTrainerWrapper
+ soup train --task unlearn.
Validated on SmolLM2-135M: ROME 0.0016->0.96, NPO/SimNPO forget loss down.
+81 tests (tests/test_v0719.py). 2 review waves, all findings fixed.
2026-06-03 17:04:03 +05:00
5652215d4a
feat(adapters,loop): v0.71.4 — live canary verdict + cmaes merge + PR push + pre-wired loop + can lineage + branch↔registry
...
Closes #172 , #173 , #176 , #177 , #220 , #223 .
- #172 soup adapters merge --canary/--strict-verdict: live OK/MINOR/MAJOR verdict (was UNKNOWN stub)
- #220 soup adapters merge --strategy cmaes: live merge→score→write-best loop (was plan-only)
- #223 soup adapters pr --push owner/repo#N: post PR comment via gh api
- #176 soup loop --pre-wired: real traces→DPO→eval-gate→canary stages
- #177 soup loop --pack-cans / replay --extract: iterations as Soup Cans + Registry lineage DAG
- #173 soup adapters branch --from-registry / --attach-to-registry
Security: backdoor-scan + license gates now run for ALL merge strategies (incl cmaes);
loop canary deploy restricted to loopback/RFC1918; gh child env from allowlist;
canary read uses O_NOFOLLOW+fstat (TOCTOU); pack-entry failure rolls back registry entry.
Tests: 12342 → 12474 (+130 in tests/test_v0714.py).
2026-06-02 12:25:01 +05:00
21a2bf8e8c
feat(governance,energy): v0.71.3 — annex PDF, audit auto-log, energy hook, can v3, airgap receipt
...
Closes #180 #181 #182 #183 #184 #188 .
- #180 EnergyTracker (codecarbon offline) + `soup train --track-energy`; [carbon] extra
- #181 PDF Annex XI/XII (reportlab) + paths.atomic_write_bytes; [pdf] extra
- #182 Soup Can manifest v3 + attestations field + `can pack --attest`
- #183 per-command audit-log auto-instrumentation (--no-audit-log / SOUP_NO_AUDIT_LOG)
- #184 auto-populate Annex top_domains from the training JSONL
- #188 embed repro-receipt into `soup airgap-bundle`
New [pdf]+[carbon] extras (reportlab also in [dev]). +83 tests (12259 -> 12342),
79.10% coverage. Reviewed (security/code/python/tdd): security M1 raw-size gate on
--attest before parse, python H1/H2 type hints, +22 negative tests.
2026-06-01 19:12:24 +05:00
9300ee3412
feat(governance,sign): v0.71.2 — ed25519 signing + supply-chain gates
...
ed25519 signing (#179/#185 ed25519 half): new utils/signing.py + [sign]
extra. `soup adapters sign/verify --backend ed25519` (--key/--generate-key/
--public-key) and `soup attest emit --sign ed25519` + new `soup attest
verify`. Sigstore keyless stays infra-blocked (OIDC/Fulcio/Rekor network).
#186 namespace-pin gate wired into download_repo (anti-AI-Jacking; fail-open)
#187 adapters merge auto-detects each adapter's license
#190 license-override reason recorded to the audit log
#191 NamespacePinStore WAL + busy_timeout + cross-process lock
#192 merge refuses scan-FAIL inputs unless --allow-unscanned
+106 tests (12153 -> 12259); full suite 79.07% cov. cryptography added to
[dev] for CI. 5 review waves, all findings fixed.
2026-06-01 17:17:44 +05:00
7724353ac6
docs: trim README to a 238-line front door; move feature reference to docs/
...
The README had grown to 5046 lines (195 sections) — roughly one deep-dive per
feature accreted over 70 releases. Split it into a concise front door plus a
public docs/ tree:
- README (5046 -> 238 lines): hero, why, quickstart, config, a Documentation
map, data formats, common commands, models, Docker, requirements, dev.
- docs/*.md: all 185 feature sections preserved verbatim, grouped into 10 themed
guides + an index. Every original line is accounted for (content-conservation
checked); all 235 internal links + anchors verified to resolve.
- un-gitignore docs/ (it was empty); fix a pre-existing dangling
docs/QUANTIZATION.md link; correct the stale `ruff check soup_cli/` ->
`src/soup_cli/` reference in the Development section.
No version bump: docs-only — rides into the 0.71.0 deps-split release.
2026-05-31 20:10:59 +05:00
Alpamys