The best DX for LLM fine tuning
Despite the number of available tools, the developer experience (DX) for LLM fine tuning still sucks. We've abstracted away its complexity to allow small and medium sized teams to focus on data, use-case, and business value.
Dataset Management
Upload custom CSV/JSONL files or sync directly from Hugging Face. We handle the formatting and tokenization.
Optimized Training
Leverage techniques like LoRA, QLoRA, and full parameter tuning with 4-bit/8-bit quantization out of the box.
Seamless Export
Export adapters, merge weights, convert to GGUF, and push directly to the Hugging Face Hub or GCS.
Instruction Tuning (SFT)
Adapt foundational models to follow precise instructions and format outputs for your specific domain.
Reasoning (GRPO)
Enhance the cognitive capabilities of Gemma for complex, multi-step tasks and logical problem-solving.
Preference Tuning (DPO)
Align your model's outputs with human preferences to improve safety, accuracy, and user experience.