Tag
Soup is an open-source CLI (Apache 2.0) that covers LLM post-training end to end: it decides whether training is worth it, writes the config, trains with LoRA, DPO or GRPO and returns a SHIP or DON'T SHIP verdict. I tested version 0.76.1 on 15 October 2026 on Linux arm64 with no GPU.
LoRA trains only two small matrices per layer and QLoRA does it on a 4-bit base model. The memory arithmetic of each technique and a reproducible example in Docker with no GPU: LoRA trained 3.5% of SmolLM2-135M's parameters with a 2.35 GB peak, against 3.88 GB for full fine-tuning.
Fine-tuning your own LLM pays off in three cases: you need a very specific style or voice, a rigid structured output format, or you want lower cost and latency from a small specialised model. LoRA and QLoRA have cut the GPU cost, but preparing data and running the model in production are still expensive. For everything else, RAG and prompt engineering are usually enough.
