bailian-finetune

About bailian-finetune

Alibaba Cloud Bailian model fine-tuning entry point: use `bl dataset` / `bl finetune` / `bl deploy` when users want to fine-tune or train their own models (fine-tune, supporting SFT / SFT-LoRA / DPO / DPO-LoRA / CPT, across text, speech, and images), validate or upload training datasets, check training progress and logs, select a checkpoint, export fine-tuned artifacts, or deploy a dedicated model as a service. The workflow is: validate data → upload to get file-id → finetune create to start a task → watch progress → export → optionally compress/quantize → deploy. An API key is required; preview write operations with `--dry-run` first. Do not trigger for fine-tuning explicitly named for Volcano Ark/ark; use bailian-model-recommend to choose a model; use bailian-gen to generate images or videos with an existing model; use bailian-cli for other Bailian resource management. Official installation: `bl skill init` (installed together with the shared protocol bailian-protocol).