Discofork
Repository brief

hiyouga/LlamaFactory

Read the upstream summary on the left, browse the cached forks below it, and load each fork comparison into the right-hand panel.

Cached analysis
cached 2026-03-30T11:51:53.203Z
3mo ago

hiyouga/LlamaFactory

LlamaFactory is an actively maintained Python project for unified, efficient fine-tuning of 100+ LLMs and VLMs. It is large and widely adopted, with 69,252 stars, 8,436 forks, recent commits on 2026-03-30, and support surfaces for CLI, Web UI, Docker, docs, examples, and tests. Forks are most likely interesting if you want to extend model support, training backends, datasets, or UI/workflow integrations in a fast-moving fine-tuning stack.

GitHub
Loading tags...
Stars69,252
Forks8,436
Default branchmain
Last pushed2026-03-30T02:47:20Z
Recommended shortcuts

Jump straight into Discofork's strongest cached fork picks, or open a compare view in one click.

Forks

Choose a fork to inspect

10 of 10 fork briefs
Selected

Choose this fork if sequence parallelism is the requirement and you can tolerate upstream lag and merge overhead. Choose upstream if you want the broadest, freshest LlamaFactory feature set and easier maintenance.

Choose this fork if you need a ChatTS-oriented training workflow and can accept divergence from upstream. Stay with upstream if you want the widest model/backend support, fresher fixes, and the full documentation/demo surface.

Choose this fork if your goal is pondering/continuous-space research and you want the paper's implementation details. Choose upstream if you need the broader, actively maintained fine-tuning platform with current backend support and fewer missing workflows.

Choose this fork only if its custom training/data behavior matches your needs and you can tolerate maintenance debt. If you want broad model support and current upstream training backend work, upstream is the safer default.

Choose this fork if you want PEFT-focused fine-tuning plus benchmark/evaluation helpers and can tolerate upstream lag. Choose upstream if you need the latest model/backend support and the broadest maintained feature set.

Prefer this fork if you are explicitly building Romanian LLM workflows and want the fork's curated data/branding/customization. Prefer upstream if you need the newest training backends, broader docs/examples, or maximum compatibility with the fast-moving LlamaFactory ecosystem.

Choose this fork if you want the added dataset package and are comfortable being behind upstream; choose upstream if you need the newest training backends, fixes, and broader model support.

Prefer this fork if your priority is long-sequence training with Ulysses and you want that capability integrated into LlamaFactory. Prefer upstream if you want current model support, broader feature coverage, and lower maintenance risk.

Prefer this fork only if you specifically need the channel-loss work and the bundled experimental assets. For general LlamaFactory adoption, upstream is the better choice because this fork is stale and materially behind on recent fixes and backend support.

Choose this fork only if you specifically want the older, simpler PEFT/QLoRA training workflow and its bundled data/fixes. For anyone starting fresh or needing current model, backend, and multimodal support, upstream LlamaFactory is the better default.