Skip to content

record(MODEL-NEMOTRON-H): the spec described a resolver that #569 replaced (#517, #569) - #582

Merged
localai-bot merged 1 commit into
mainfrom
row/MODEL-NEMOTRON-H-SPEC-RECONCILE
Aug 13, 2026
Merged

record(MODEL-NEMOTRON-H): the spec described a resolver that #569 replaced (#517, #569)#582
localai-bot merged 1 commit into
mainfrom
row/MODEL-NEMOTRON-H-SPEC-RECONCILE

Conversation

@localai-bot

Copy link
Copy Markdown
Collaborator

A fresh review of #569 found .agents/specs/nemotron-h-model.md asserting something the tree no longer does: the W1 LOW-3(b) section claimed the resolver "gates the CHECKPOINT_ROOT path on that manifest", and its evidence row B (add 29f2d174….json, nothing else changed → EXIT=0) no longer reproduces.

Worse than stale — it documents a proof by construction whose construction is void, so a reader would take it as evidence the pin works as described.

Records only. agent-preflight.sh --staged green.

🤖 Generated with Claude Code

…laced (#517, #569)

FOLLOWING_AGENTS_PROTOCOL

A fresh review of #569 found this spec asserting something the tree no longer
does. The W1 LOW-3(b) section said Nemotron35LightningSnapshot() "gates the
CHECKPOINT_ROOT path on that manifest", and its evidence row B -- add
29f2d174....json, nothing else changed, EXIT=0 -- does not reproduce any more.
Adding the manifest now changes nothing; the per-file download sidecars decide.

Records cannot disagree with the tree, and this one is worse than merely stale:
it documents a proof-by-construction whose construction is void, so a reader
would take it as evidence the pin works the way it says.

The section is marked SUPERSEDED with the reason -- the manifest records "this
revision was downloaded here once", not "these bytes are that revision", and
huggingface_hub's _tree_cache.py never invalidates, so manifests ACCUMULATE and
an old revision's manifest vouches for new bytes. Row B is retained rather than
deleted, labelled as no longer true, because what was tried and why it was
insufficient is the part a future porter needs.

Also links the two gate-infrastructure issues this row depends on, closing the
review's second finding: the protocol wants an issue linked from the roadmap
table, the row's spec and the PR, and #569 was in only one of the three. The
roadmap table half landed in ce70711.

Carried into the spec because it is an operating condition and not a defect:
W6 must run with VT_NEMOTRON35_SNAPSHOT UNSET and must record the resolved
directory, since that override is deliberately never revision-checked and would
silently accept a substituted checkpoint -- which is exactly what a token gate
cannot see.

Records only, no product code touched.

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:claude-opus-5 [Claude Code]
@localai-bot
localai-bot merged commit 65625f3 into main Aug 13, 2026
16 of 18 checks passed
localai-bot pushed a commit that referenced this pull request Aug 13, 2026
…ver correction (#517)

Fifth re-merge; `origin/main` `751325460` -> `65625f34a` (#582), a records-only
commit that rewrites §2 of the SAME keyed record this branch also edits in §2.
Source and tests are untouched by it, so no gate is invalidated.

`.agents/specs/nemotron-h-model.md` was merged BY HAND again per AGENTS.md:
`git checkout origin/main -- <spec>` first, discarding git's automatic
resolution, then the three W2 regions re-applied with uniqueness-asserted
anchors. Verified block by block (md5 per `##` heading) -- **16 blocks, 13
BYTE-IDENTICAL to main, 3 differ and all three are W2's own**: §2's single table
row (`apply_routed_scale_to_output=True` at `nemotron_h.py:234`, factor `:233`),
the new §6a, and §7 `Now`. Main's rewritten §2 PROSE is carried whole -- the W2
edit is one row of the anchor table above it and does not touch the resolver
narrative #582 corrected.

Delta vs `origin/main` remains exactly this branch's eight W2 files.

FOLLOWING_AGENTS_PROTOCOL

Refs #517. Refs #582.

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:claude-opus-5 [Claude Code]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants