Submit Job
Single-node training only. Jobs run on a single EC2 instance with up to 8 GPUs. Multi-GPU strategies such as DeepSpeed, FSDP, tensor parallelism, and sequence parallelism are all supported. Only multi-node Ray configs (use_ray: true, ray_num_workers > 1) will be rejected.
Upload a file or provide a dataset path
Drag & drop your dataset, or click to browse
JSONL, JSON, CSV, Parquet · max 5 GB
HuggingFace ID, S3 URI, or public URL
Axolotl-format YAML
Drag & drop your YAML config, or click to browse
Quick start examples:
Generate or upload a config to continue