https://guijiagi.com/categories/%E5%BC%80%E6%BA%90/Recent content in 开源 on AI 实战派 · 从技术到赚钱Hugozh-cn本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权Wed, 07 Oct 2026 17:18:00 +0800https://guijiagi.com/posts/2026-10-07-llama-factory-finetuning/Wed, 07 Oct 2026 17:18:00 +0800https://guijiagi.com/posts/2026-10-07-llama-factory-finetuning/LLaMA-Factory 把 LoRA、QLoRA、全参微调压进一个 Web UI。从数据集格式到训练参数，讲清一次微调怎么落地。https://guijiagi.com/posts/2026-10-07-vllm-paged-attention-serving/Wed, 07 Oct 2026 17:06:00 +0800https://guijiagi.com/posts/2026-10-07-vllm-paged-attention-serving/从 PagedAttention 到连续批处理，再到 2026 年的 Model Runner V2，vLLM 是开源推理服务的吞吐标杆。拆解它的调度内核。https://guijiagi.com/posts/2026-10-07-llama-cpp-inference-engine/Wed, 07 Oct 2026 17:00:00 +0800https://guijiagi.com/posts/2026-10-07-llama-cpp-inference-engine/从 GGUF 量化到 MTP 投机解码，llama.cpp v0.6.0 把 320B MoE 拉进了消费级显存。拆解它的工程内核与适用边界。