Brev.dev Notebooks

This repo contains helpful AI/ML notebook templates for LLMs, multi-modal models, image segmentation, and more. Each notebook has been coupled with the minimum GPU specs required to use them along with a 1-click deploy badge that starts each notebook on a GPU.

Contributing

We welcome contributions to this repository! If you have a notebook you'd like to add, please reach out to use the Discord or open a pull request!

Notebooks

We've split the notebooks into categories based on the type of task they're designed for. Our current split is: LLM finetuning/training, multi-modal models, computer vision/image segmentation, and miscellaneous. Let us know if you want to see more notebooks for a certain task or using different frameworks and tools!

LLM Finetuning/Training

Notebook	Description	Min. GPU	Deploy
Fine-tune Llama3 using Direct Preference Optimization	Fine-tune Llama3 using DPO	1x A100
Fine-tune Llama3 using SFT	Fine-tune and deploy Llama 3	2x A100
Fine-tune Llama 2	A Guide to Fine-tuning Llama 2	1x A10G
Fine-tune Llama 2 - Own Data	Fine-tune Llama 2 on your own dataset	1x A10G
Fine-tune Mistral	A Guide to Fine-tuning Mistral	1x A10G
Fine-tune Mistral using NVIDIA NeMO and PEFT	Fine-tune Mistral using NVIDIA NeMO and PEFT	1x A100
Fine-tune Mistral - Own Data	Fine-tune Mistral on your own dataset	1x A10G
Fine-tune Mixtral (8x7B MoE)	A Guide to Fine-tuning Mixtral, Mistral's 8x7B MoE	4x T4
Fine-tune Mixtral (8x7B MoE) - Own Data	A Guide to Fine-tuning Mixtral on your own dataset	4x T4
Fine-tune BioMistral	A Guide to Fine-tuning BioMistral	1x A10G
Fine-tune Phi-2	A Guide to Fine-tuning Phi-2	1x A10G
Fine-tune Phi-2 - Own Data	Fine-tune Phi-2 on your own dataset	1x A10G
Training Question/Answer models using NVIDIA NeMo	Use NeMo to train BERT, GPT, and S2S models for Q&A tasks	1x A10G

LLM Inference/Deployment

Notebook	Description	Min. GPU	Deploy
Run inference on Llama3 using TensorRT-LLM	Run inference on Llama3 using TensorRT-LLM	1x A10G
Inference on DBRX with VLLM and Gradio	Run inference on DBRX with VLLM and Gradio	4x A100
Run BioMistral	Run BioMistral	1x A10G
Run Llama 2 70B	Run Llama 2 70B, or any Llama 2 Model	4x T4
Use TensorRT-LLM with Mistral	Use NVIDIA TensorRT engine to run inference on Mistral-7B	1x A10G
StreamingLLM for Optimized Inference	Use StreamingLLM for infinite length input without finetuning	1x A10G

Multi-modal and Computer Vision Models

Notebook	Description	Min. GPU	Deploy
Finetune and deploy LlaVA	Finetune the LlaVA model on your own data	1x A10G
AUTOMATIC1111 Stable Diffusion WebUI	Run Stable Diffusion WebUI, AUTOMATIC1111	1x A10G
ControlNet on AUTOMATIC1111	Run ControlNet Models on Stable Diffusion WebUI	1x A10G
SDXL inference with LoRA and Diffusers	Run inference using LoRA adaptors and SDXL	1x A10G
Oobabooga LLM WebUI	Run Oobabooga, the LLM WebUI (like AUTOMATIC1111)	1x A10G
EfficientViT Segement Anything	Run a TensorRT optimized version of Segment Anything	1x A10G

Other applications and tools

Notebook	Description	Min. GPU	Deploy
Deploy to Replicate	Deploy Model to Replicate	any || CPU
GGUF Export FT Model	Export your fine-tuned model to GGUF	1x A10G
Julia Install	Easily Install Julia + Notebooks	any || CPU
PDF Chatbot (OCR)	PDF Chatbot using OCR	1x A10G
Zephyr Chatbot	Chatbot with Open Source Models	1x A10G
Accelerate Data Science using NVIDIA RAPIDS	Accelerate Data Science using NVIDIA RAPIDS	1x A10G

---

What is Brev.dev?

Brev is a dev tool that makes it really easy to code on a GPU in the cloud. Brev does 3 things: provision, configure, and connect.

Provision:

Brev provisions a GPU for you. You don't have to worry about setting up a cloud account. We have solid GPU supply, but if you do have AWS or GCP, you can link them.

Configure:

Brev configures your GPU with the right drivers and libraries. Use our open source tool Verb to point and click the right python and CUDA versions.

Connect:

Brev.dev CLI automatically edits your ssh config so you can ssh gpu-name or run brev open gpu-name to open VS Code to the remote machine