Add fine-tuning service (no SDK)#29

Open

fingriffin wants to merge 33 commits into

mainfrom

feature/finetuning-service

Commits on Apr 20, 2026

feat(docs): add initial draft ADR for fine-tuning service
Fin Griffin
committed

Commits on Apr 22, 2026

feat(docs): update ADR-011 with SDK decision
Fin Griffin
committed
refactor(docs): create dedicated ADRs folder for fine-tuning service
Fin Griffin
committed
chore(docs): remove mention of SDK ADR
Fin Griffin
committed
feat(stacks): add lightweight docker compose for fine-tuning service
Fin Griffin
committed
feat(stacks): add Dockerfile for fine-tuning service build using axolotl base image
Fin Griffin
committed
feat(stacks): add fine-tuning service FastAPI application
Fin Griffin
committed
feat(docs): add ADR for fine-tuning service skeleton
Fin Griffin
committed
Merge pull request #26 from acceleratescience/feature/ft-service-skeleton
Fin Griffin
authored
chore: fix typo on endpoint name
Fin Griffin
committed

Commits on Apr 23, 2026

chore: migrate to hardcoded axolotl base image
Fin Griffin
committed
fix(docker): override system python guard for requirements installation
Fin Griffin
committed
feat(stacks): fallback to plain python axolotl base image
Fin Griffin
committed
feat(stacks): add lightweight user config for fine-tuning service and pre-axolotl validation
Fin Griffin
committed
feat(stacks): add axolotl subprocess with separate worker
Fin Griffin
committed
chore: add chat template to test dataset and initialise db from axolotl worker
Fin Griffin
committed

Commits on Apr 24, 2026

feat(stacks): let device for fine-tuning service be a environment variable
Fin Griffin
committed
feat(deps): add FA4
Fin Griffin
committed
fix: pin FA4 version with cu13 wheels, increase default sequence_len, add sample packing (training only)
Fin Griffin
committed

Commits on Apr 27, 2026

feat(stacks): add do_eval as key in fine-tuning service
Fin Griffin
committed
feat(stacks): add wandb tracking
Fin Griffin
committed
feat(docs): add ADR for axolotl implementation
Fin Griffin
committed
Merge pull request #27 from acceleratescience/feature/axolotl
Fin Griffin
authored
feat(stacks): add auth layer to FastAPI for fine-tuning service
Fin Griffin
committed

Commits on Apr 28, 2026

feat(stacks): add docker network between llm and fine-tuning service
Fin Griffin
committed
feat(stacks): add user id whitelist for fine-tuning service
Fin Griffin
committed
fix: migrate to gitignored whitelist for user auth
Fin Griffin
committed
fix: fix lite llm port typo
Fin Griffin
committed
Merge pull request #28 from acceleratescience/feature/auth
Fin Griffin
authored

Commits on Apr 29, 2026

feat(docs): add ADR for authentication
Fin Griffin
committed
chore(docs): fix typo in adr title
Fin Griffin
committed
feat(nginx): add fine-tuning service to nginx config
Fin Griffin
committed

Commits on May 1, 2026

fix: use python.urllib for health check
Fin Griffin
committed