Soup CLI can fine-tune Llama 3.1 8B on a 4 GB RTX 3050 Laptop GPU. It controls VRAM with streamed base weights and NF4 quantization, supporting LoRA, full fine-tuning, DPO, ORPO, SimPO, KTO, resume, and NVMe tiering.
Features
- 8B tuning on a 4 GB GPU
- Streamed weights and NF4
- Preference-alignment methods
- Resume and VRAM preflight
Use Cases
- Laptop fine-tuning
- LoRA adapter training
- Preference alignment
- Low-VRAM experiments
FAQ
Soup CLI can fine-tune Llama 3.1 8B on a 4 GB RTX 3050 Laptop GPU. It controls VRAM with streamed base weights and NF4 quantization, supporting LoRA, full fine-tuning, DPO, ORPO, SimPO, KTO, resume, and NVMe tiering. Core capabilities include: 8B tuning on a 4 GB GPU, Streamed weights and NF4, Preference-alignment methods.
Common scenarios include: Laptop fine-tuning, LoRA adapter training, Preference alignment.
