Parallel teach lifecycle — source snapshot 2026-09-12

Source-only prerelease of matching core, node, and CLI changes. Core exposes an operator-configurable active-jobs-per-key limit; node enforces it with a 70-job concurrent admission regression test; CLI adds ainize teach parallel, using ain-js for finalized per-job status paths confirmed by an independent validator. The default per-key limit remains 2. GPU trainer slots are unchanged: 70 admitted/queued jobs do not prove 70 simultaneous GPU training processes. Real 70-job on-chain end-to-end evidence is not claimed by this source release. This is not an npm package publication. CLI usage: https://github.com/ainblockchain/ainize-cli/blob/teach-parallel-20260912/docs/parallel-teach.md

Experimental: Hugging Face URL dataset import

Existing Hugging Face datasets → Ainize teaching

Experimental source release of ainize dataset <huggingface-url>, commit 3311573afb93c7e6f0877a864c16b104c03efcfb.

• Imports an existing HF dataset into the selected Ainize node through the signed teaching-dataset API. No new HF repository is created and nothing is published to the Hub.
• Supports repository, viewer and resolve URLs, revision binding, config/split selection, explicit row limits, scalar column mapping and immutable file fallback when the viewer is unavailable.
• Records private input/upload provenance and returns the real Ainize dataset ID. Does not automatically train, consent to publication, bypass validation or create marketplace listings.
• Continues through the existing dataset → teach train → teach publish → use/chat commands. Public catalog presence, independent LISTED verification, payment, model loading and answer accuracy are distinct outcomes. Same-ledger peer connectivity and a reachable seller endpoint are needed; no production ledger is silently re-initialized.

Validation and limits

• Docker: 2 CPUs, cpuset 0–7, 4 GiB memory, no additional swap, read-only filesystem, isolated network. All 56 CLI tests passed, including 11 new HF tests. Tests use fixture data and a local stub node where applicable; they do not establish GPU training success.
• Tested image: sha256:563e96f6725136939bd6bac5cf6a6480fcc535f71fb51c0ab02ab19d7d8891d7 (local image, not a public registry publication).
• Real external HF lhoestq/demo1 train split: 5 rows imported with review/star mapping. Existing DART canonical file: 8 rows imported with the original dataset ID and hash, created=false, no new job. These are integration checks, not 100 trained models or marketplace sales.
• Import size is limited to 32 MiB; viewer imports above 10,000 rows require explicit sampling. Only data convertible to Ainize teaching rows is supported. Node row/PII/format limits still apply. Gated access terms and data reuse rights are not waived.
• The first real custom-column import failed because the node parser ignored JSONL column mappings. Mapping now occurs in the adapter, retaining both input and upload hashes. A build-context contamination from old untracked tests was fixed by clearing only source/test/dist paths inside the new Docker image before copying the selected source. Failed evidence was retained.

Installable release asset

The attached ainize-0.1.0.tgz is built from the same source commit and tested image. It was installed into an empty Docker directory with dependencies from the public npm registry, then used for a real HF-to-Ainize import. The existing dataset ID/hash was preserved, with no new training job. The public release download was fetched without authentication and is byte-identical to that tested package.

Node.js 24 or later is required. This installs the CLI; a configured Ainize node and teaching identity remain prerequisites for importing datasets.

npm install -g https://github.com/ainblockchain/ainize-cli/releases/download/year3-hf-dataset-import-20260911/ainize-0.1.0.tgz

SHA256: a4faae27db9d256b511e67df9432bc89ba321adfe65c62a0d28361e5dcc5a9bf. A checksum file is attached. The attempted npm registry publication as ainize@0.1.0, tag hf-integration, was rejected with EOTP: maintainer two-factor authentication is required and was not bypassed. npm install ainize is therefore not a verified installation route for this release.

This publishes Git source, test evidence and an installable GitHub release asset, not a new npm registry version, HF dataset/model/Space, Docker registry image or public Ainize knowledge listing. Existing npm releases do not gain this feature merely because this release exists. See the repository README for supported commands and prerequisites.

@ainize/cli 0.1.0

First public release of the ainize CLI.

npm install -g @ainize/cli

• Package is now public on npm (MIT); @ainize/core and @ainize/node (optional peer) resolved from the registry at ^0.1.0; engines.node >= 24.