中文 README | Documentation index
Cinematic AI training logs for demos, tests, teaching, and suspiciously serious terminals.
TrainFake is a preset-first terminal simulator for believable AI training runs. It does not train a real model, download weights, or touch a GPU. It stages the terminal experience around environment scans, warmup, training loops, incidents, recovery logs, eval-like metrics, checkpoints, and summaries.
Python 3.8 or newer is required.
python3 -m pip install .
trainfake presets
trainfake run boss-is-watching --step-delay 0
trainfake tui boss-is-watchingRun a custom YAML script:
trainfake run --config tests/fixtures/sample_run.yaml- Built-in scenes for LLM pretraining, Diffusers LoRA, vLLM launch rehearsals, BERT fine-tuning, RLHF reward models, RAG evals, multimodal pretraining, and GPU cluster drills.
- Classic streaming logs with
trainfake run. - Split-screen Textual dashboard with
trainfake tui. - YAML run scripts for custom stages, events, log styles, checkpoint cadence, and reproducible seeds.
MIT