Skip to content

docs(kolibri1): document CPU chat and current limits - #3451

Open
localai-org-maint-bot wants to merge 3 commits into
localai-org:mainfrom
localai-org-maint-bot:docs/source-alignment-20261012
Open

localai-org-maint-bot wants to merge 3 commits into
localai-org:mainfrom
localai-org-maint-bot:docs/source-alignment-20261012

Conversation

@localai-org-maint-bot

Copy link
Copy Markdown
Collaborator

Row

DOCS-KOLIBRI1-CURRENT-USAGE, a documentation-only work ID. No model or backend lifecycle changes.
Canonical issue: .agents/issues/_owed/ISSUE-LOCAL-01M4Q12F0EMYDYQV1P1PNJKEB3.md.

Before starting

  • Started from upstream main at c6a971e53 in a separate worktree.
  • No open documentation claim covered this correction. The existing CPU and TT implementation issues retain their scope.
  • One PR carries the spec and implementation, following the repository default. Spec commit: 23a85fa47.
  • Inspected the Kolibri registry, loader, forward, dequantization cache, parser registrations, executable definitions, and retained CPU and serving evidence.
  • No matrix row changes or helper-readiness gate apply to this editorial correction.

What changed

The feature row described an obsolete loader and called the Tenstorrent arm owed. Replace that history with current behavior and validation limits. Add October news and CPU completion/chat recipes with explicit parser selection. Preserve the incomplete checkpoint provenance and distinguish partial CPU token agreement from unverified author-plugin parity.

Evidence

  • Full scripts/agent-preflight.sh passes. bash scripts/agent-preflight.sh --staged exited 1 with the same 20 failing checks as the baseline, plus 13 skips. Missing PyYAML, CMake, Ninja, C++ tools, readelf, and NumPy limit this environment; isolated-Python fixtures also fail. Full preflight is not green.
  • check-readme-structure.py, check-supported-models.py, check-surface-coverage.py, check-benchmark-index.py, and check-quickstart-recipes.py under scripts/: all pass with python3.
  • Corresponding tests/scripts/test_check_*.py suites: 117 tests pass.
  • check-agent-record.py, check-site.py, commit trailer/style checks, and git diff --check: pass.
  • python3 scripts/check-tree-compiles.py --base c6a971e53: no C++ source, header, or build file is in scope.
  • Public edits are limited to the Kolibri news entry, feature row, and usage section. Commands, links, and claims were checked against source.

Independent review passed on ee4aecfc1. A scratch mutation removing the README Features heading failed its guard as expected. Source and command validation used no GPU.

Speed claims

  • This PR makes no new speed claim and changes no benchmark disposition.

Honest gaps

No model execution, GPU work, or model download was performed. The commands are checked against source, not a local 78.8 GB checkpoint. Full correctness and speed parity against the author's vLLM plugin remain unverified. The source record supplies only a short checkpoint revision and lacks shard hashes. The canonical documentation issue remains open pending upstream landing.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: Codex:GPT-6 [Codex]

The feature row describes an obsolete loader and omits the reachable TT
path. Define source checks and CPU-only gates before updating public docs.

Tracks ISSUE-LOCAL-01M4Q12F0EMYDYQV1P1PNJKEB3.
No runtime behavior or model lifecycle changes. No GPU gate applies.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: Codex:GPT-6 [Codex]
The public row described an obsolete loader and called the TT arm owed.
Document the reachable paths, CPU commands, and parser selection.
Retain the incomplete artifact provenance and partial token agreement.

Resolve the documentation gap in ISSUE-LOCAL-01M4Q12F0EMYDYQV1P1PNJKEB3.
Its local issue stays open pending review and upstream landing.
Five scoped checkers and 117 mutation tests pass. Full staged preflight
is pending with known missing-tool failures recorded in the spec.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: Codex:GPT-6 [OpenAI]
The checkpoint index records tensor data size, not total shard file size.
Name that distinction in the usage guide and record the review evidence.

Issue: ISSUE-LOCAL-01M4Q12F0EMYDYQV1P1PNJKEB3

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: Codex:GPT-6 [Codex]

This branch has not been deployed

No deployments
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.

1 participant