v0.1.0: DomainPostTrain initial release

Highlights

• General domain post-training pipeline: CPT -> Fact-SFT -> optional DPO -> merge -> post-training quality evaluation.
• Static AsterHelp mock data for CPT, Fact-SFT, DPO, and quality evaluation examples.
• Qwen3.5 4B default config with LoRA/QLoRA-oriented training paths.
• GGUF export from the merged model as the default deployment route; ONNX export kept optional through requirements-onnx.txt.
• OpenAI-compatible Flask inference service with OpenAPI documentation endpoint.
• GitHub Actions CI, Issue Templates, PR Template, and Discussion category forms for community collaboration.
• README architecture diagram for the end-to-end data and model artifact flow.

Validation

• py -3.10 -m compileall -q pipeline scripts serve_inference.py
• JSONL schema checks for data/sft, data/dpo, and data/eval
• README/config reference consistency checks
• Legacy domain keyword scan across public project files
• GitHub Actions CI passed on main