VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
u_Unrolled
All projects

LLMs · Intermediate · by Unsloth AI

Unsloth: 2–5× faster LLM fine-tuning

Free, open-source fine-tuning framework that makes LoRA/QLoRA 2–5× faster with 80% less memory via hand-optimized CUDA kernels — runs on consumer GPUs.

1–2 weekends 79500 on GitHubPythonPyTorchCUDATransformersbitsandbytes
Open the code on GitHub

What you'll learn

  • Parameter-efficient fine-tuning (LoRA/QLoRA) from scratch
  • Memory-efficient 4-bit quantization for consumer hardware
  • Custom dataset preparation and chat template formatting
  • Exporting fine-tuned models to GGUF for local inference

Read first

The lessons that explain what this project uses.

Part of the LLMs in Depth path.

Milestones

0/4

Built it? Show it.

Add your own repo to earn 150 XP and the Shipper badge. It appears on your profile, so it's ready for your resume.

More llms projects