mirror of
https://github.com/axolotl-ai-cloud/axolotl.git
synced 2026-10-12 02:11:20 +00:00
Wing Lian
f5dec2b5eb
* docs: offline and air-gapped deployment guide with bundle script Add docs/airgapped.qmd covering a two-host offline install: stage the pinned repo, uv, a managed Python 3.14 interpreter, a cu130/torch 2.14 wheelhouse for the target platform (incl. fla, ringmaster, ray, deepspeed and the cut-cross-entropy plugin), the HF cache with model, dataset and flash-attn hub kernels pinned to the transformers kernel version table, then bundle, verify, install offline and train with DO_NOT_TRACK and HF offline mode set. scripts/airgap/build_bundle.sh automates the staging half. Link the guide from the installation page and the Deployments sidebar. * fix(airgap): make WITH_CCE=0 and reruns work in build_bundle.sh Pass --clear to uv venv so rerunning a step does not fail on an existing environment. Only install and record the cut-cross-entropy wheel when WITH_CCE=1, and avoid a non-zero function return under set -e when it is disabled. * docs(airgap): move Quick reference up to follow Prerequisites Readers who want to start quickly should not have to pass the layout, variables and kernel sections first. The lead-in now points forward to those sections instead of assuming they were read.