Announcing Lepton AI open beta! Industrial-strength AI Inference with just a few lines of Python code – with automatic performance and reliability matching the hyperscalers. Forget docker and other infra nuts and bolts – we’ve got you covered!
More at:
We are super excited to partner with
@SambaNovaAI
to server the most state-of-the-art Samba-CoE-v0.3 model. In the LLM land, innovation happens with a joint effort on the hardware side and software side. 🧵
🚀 Excited to introduce the future of machine learning on Lepton AI! We've teamed up with Google's latest marvel, Gemma, to bring you an API that's as powerful as it is user-friendly. 🌟 Start exploring with Gemma today at .
#MachineLearning
#AI
#Gemma
We transparently support PyTorch, HuggingFace Transformer, and other common AI libraries at the base. We also work closely with awesome open source libraries like vLLM - in fact, launching vLLM has never been easier.
🚀 Introducing Illusion Diffusion Model Apis by Lepton AI
Transform your wildest creative ideas into mesmerizing visuals! 🎨✨ Dive into a world where imagination meets reality.
#IllusionDiffusion
#UnleashCreativity
On the first Friday of 2024😝😝😝, we'd love to present you with our latest progress on OSS LLM models: Structural Decoding (Function Calling) for all Open LLMs!
Happy Friday! 🥳
🚀 Introducing "Prompt Craft" made with PromptLLM from
@hippoml_com
! Perfect simple prompts with ease to unleash creativity and efficiency like never before. 🌟💡🤖
Can you tell which one is made by AI? 😝😝😝
#LLM
#Promptshare
#promptllm
Let’s share some numbers. On LLaMA, our optimized runtime runs up to 90 times faster than the transformer baseline, and is available at today. Don't worry about optimization - we incorporated all for you.
Lepton is created by a group of passionate open source contributors behind caffe, onnx, pytorch 1.0, etcd and more. We’ve open-sourced substantial portions of our stack, and will continue to open-source more in the upcoming months. Stay tuned!
Guess who got a spot on HuggingFace Space of the Week? It's us! Try to checkout the emoji re-imagined with LCM 🚀🚀🚀
Bet you've never seen an emoji like this 😆
#huggingface
#lcm
#SDXLturbo
#sdxl
#imageGen
Container technology is great in standardizing deployment. But many developers (ourselves included) don’t need as many details as docker exposes. We allow you to specify your dependencies purely in Python. You’ll be free from infra details most of the time!
🚀 Lepton AI's model API now supports
#NodeJS
calls!
🖥️ Build smarter and faster with the power of AI in your favorite runtime.
Dive in now!
#LeptonAIUpdate
@SambaNovaAI
SambaNova's hardware architecture enables one to deliver models with composition of experts - a much interesting modeling approach on top of existing awesome architectures to deliver state of the art results on benchmarks like Alpaca. Check it out at
We introduce a Python class called Photon that bridges ML to FastAPI and does magic underneath. Implementing this subclass for your model is all it takes – we’ve wrapped several GenAI models (from HuggingFace etc.) such as Whisper, LLaMa and SegmentAnything.
We've also enabled Medusa for LLM models like Llama2 at
- Llama2-7b :
- Llama2-13b :
- Llama2-70b :
Let us know what other models you'd like to try Medusa with! 🖖
(2/2)
@OpenAI
just released fine-tuning capability today. Do you know that you can use a simple JSON file to kick off training a fine-tuned model? You'll also have the ability to do dedicated deployment with the fine-tuned model. Let us know at info
@lepton
.ai if you are interested!
Medusa is probably one of the most elegant accelerated inference solution we have seen over the last year. It runs complementary to other numerical ones (like int8/fp8, compilation etc) and gives something around ~2x performance gain in practice.
(1/x)
As a result, for many HuggingFace models, you can simply launch it in one line by specifying `hf:model_name`. This includes text generation, image classification, sentiment analysis, and more to come.
🚀Check out the blog post by
@YuzeMa5
for the story and tip on fine-tuning open-source models with
@LeptonAI
and
@LangChainAI
. Let us know what you think!
lots of folks are talking about how best to finetune an open-source model for their specific use case–
@YuzeMa5
@LeptonAI
*actually* did it
more on their process and learnings
@a13935257451
@jiayq
We do have APIs where you can call with your workspace token (Default access will be rate limited, but you can let us know at info
@lepton
.ai for unlimited access)