LlamaFactory tested by Argusic Runner

What the test found

LLaMA-Factory 0.9.6.dev0 is fully installed and functional: the Python package imports cleanly, all pytest test suites pass using real tiny models from HuggingFace (training, inference, chat, export, data processing, template parsing, model loading), and the CLI tools (llamafactory-cli/lmf) with env, version, train, chat, export, webchat, webui, and API server commands all work.

No errors encountered. All tests used real HuggingFace tiny-random models from llamafactory namespace. Tested: pip bootstrap (no system apt), pip install of all deps, editable install of llamafactory. Ran 129/137 data tests, 57/61 model tests, 58/64 train tests, 2/2 eval tests, 7/7 e2e real training tests (pt/sft/dpo/kto/rm), 2/2 chat tests, 71/81 tests_v1 tests. All passed with only expected skips for gated models and GPU-only distributed tests. CLI (llamafactory-cli and lmf) confirmed working.

Argusic installed LlamaFactory in 8.5 minutes and launched it; verification reached Verified running for real.

Runsall runs of LlamaFactory

This summary is drawn from the agent's recorded report for this run. Every figure traces to the log and recording above; nothing here is authored.