New Pipeline

Configure a fine-tuning job step by step

ProjectWhich project does this pipeline belong to?

Pipeline Name

Description

Base ModelQwen/Qwen3-14B

Training MethodFull-precision LoRA adapters. Best balance of quality and efficiency.

HuggingFace Dataset IDe.g. username/dataset-name or organization/dataset-name

Dataset Format

SplitHuggingFace dataset split to use for training

The dataset will be streamed from HuggingFace at training time. Make sure the dataset is public or your HF token has access.

For instruction-tuning (qlora/lora/full_sft), your dataset should contain instruction and output fields, or be in ShareGPT/ChatML format.

Hardware

Estimated cost

$1.50/hr

Hardware

A10G Large (24GB)

Output Model NameName for the saved adapter/model checkpoint

Push to HuggingFace Hub

Automatically upload the trained adapter to your HF account after training completes.