ellora

New Ellora recipe: invent a dataset from one sentence with DataInventor (200 banking support chats), then LoRA-tune Qwen2.5-1.5B on a free Colab T4 in 3.4 minutes. Held-out perplexity 5.72 to 4.35. Open tools end to end. codelion/ellora

Adaption Labs runs a "Zero to Fine-Tuned" session today. Same two steps with open tools, as a free Colab: DataInventor writes 200 banking support chats from one sentence, then a LoRA on Qwen2.5-1.5B trains in 3.4 min. Held-out perplexity 5.72 to 4.35. codelion/ellora

@jackson_stokes Parts of that loop are already free. DataInventor generates and curates the training data, and Ellora has recipes for training a model on it with LoRA. huggingface.co codelion/ellora

@christophersaum The open-source pieces for this already exist. Ellora has recipes for teaching a model a new skill with LoRA, and DataInventor builds the training data: invent rows from a description, or curate the documents you already have. codelion/ellora huggingface.co