serving
lit-llama
Our great sponsors
serving | lit-llama | |
---|---|---|
12 | 23 | |
6,071 | 5,789 | |
0.2% | 2.6% | |
9.8 | 8.4 | |
3 days ago | 4 months ago | |
C++ | Python | |
Apache License 2.0 | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
serving
-
Llama.cpp: Full CUDA GPU Acceleration
Yet another TEDIOUS BATTLE: Python vs. C++/C stack.
This project gained popularity due to the HIGH DEMAND for running large models with 1B+ parameters, like `llama`. Python dominates the interface and training ecosystem, but prior to llama.cpp, non-ML professionals showed little interest in a fast C++ interface library. While existing solutions like tensorflow-serving [1] in C++ were sufficiently fast with GPU support, llama.cpp took the initiative to optimize for CPU and trim unnecessary code, essentially code-golfing and sacrificing some algorithm correctness for improved performance, which isn't favored by "ML research".
NOTE: In my opinion, a true pioneer was DarkNet, which implemented the YOLO model series and significantly outperformed others [2]. Same trick basically like llama.cpp
[1] https://github.com/tensorflow/serving
-
[D] How do OpenAI and other companies manage to have real-time inference on model with billions of parameters over an API?
I mean, probably - it's written in C++ https://github.com/tensorflow/serving
-
Should I wait for the M2 Macbook Pro?
We’re looking into that solution at the moment, the issue I’m referring to is related to this https://github.com/tensorflow/serving/issues/1948 we’ll know if the plug-in approach works for our uses soon but haven’t started looking into implementing it yet
- TF Serving has been unavailable for 9 days so far due to outdated GPG key
- TF Serving has been unavailable for 8 days
-
Would you use maturin for ML model serving?
Which ML framework do you use? Tensorflow has https://github.com/tensorflow/serving. You could also use the Rust bindings to load a saved model and expose it using one of the Rust HTTP servers. It doesn't matter whether you trained your model in Python as long as you export its saved model.
-
Is LaMDA Sentient? – An Interview [pdf]
Most likely it's a model server running something like https://github.com/tensorflow/serving and if there isn't a lot of load, the resource could kill some of its tasks. I wouldn't imagine it's sitting around pondering deep thoughts.
-
Ask HN: How to deploy a TensorFlow model for access through an HTTP endpoint?
https://github.com/tensorflow/serving
https://thenewstack.io/tutorial-deploying-tensorflow-models-...
-
Popular Machine Learning Deployment Tools
GitHub
-
If data science uses a lot of computational power, then why is python the most used programming language?
You serve models via https://www.tensorflow.org/tfx/guide/serving which is written entirely in C++ (https://github.com/tensorflow/serving/tree/master/tensorflow_serving/model_servers), no Python on the serving path or in the shipped product.
lit-llama
-
LLMs on your local Computer (Part 1)
A pure Python library for running several open source LLMs locally, such as Mistral or StableLM. It also branched into subprojects for supporting specific LLM model-families, such as lit-llama. The projects goal are simplicity and optimization to run on several hardware. When used with 4bit quantization, it needs below 1GB to run a 3B model, and also CPU usage remains low.
-
IT Veteran... why am I struggling with all of this?
For training: This is just an inherently technical process but looking at existing workflows helps pull the piece together and here is a good resource I got started with (https://github.com/philschmid/deep-learning-pytorch-huggingface). There is also a ton of resources on Github for scripts, walkthroughs and more (eg: https://github.com/Lightning-AI/lit-llama/blob/main/howto/train_redpajama.md)
-
Finetuning with HuggingFace
Try this, very easy to follow and reproduce the results: https://github.com/lightning-AI/lit-llama
- Llama.cpp: Full CUDA GPU Acceleration
-
What are other transformer python projects like Karpathy's nano-gpt [Discussion]
If you liked Karpathy's nano-gpt, you could checkout Lit-LLama, which is a Pytorch Lightning's reimplementation of LLama models, based on nano-gpt.
-
March 2023
Independent implementation of LLaMA that is fully open source under the Apache 2.0 license (https://github.com/Lightning-AI/lit-llama)
22-Mar-2023 Adobe unveils creative generative AI model, Firefly, to aid content creation Google has begun rolling out early access to its Bard chatbot in the US and UK Data Breach At ChatGPT? Users Report Seeing Unknown Conversations On Their Screens GPT-4 is available in preview in Azure OpenAI Service AI-powered coding assistance REPL that pairs GPT-4 (https://github.com/jiggy-ai/pair) Open source alternative to ChatGPT (https://github.com/nichtdax/awesome-totally-open-chatgpt) Run 100B+ language models at home, BitTorrent‑style (https://petals.ml/) Find the most relevant piece of code context. Hover and highlight blocks of code, the tool will point you to the most relevant pieces of information on git, messaging, and ticketing systems. Finally, it provide a summary with the power of GPT.(https://www.watermelontools.com/) Why AI Won't Replace Software Engineers (https://softwarecomplexity.com/why-ai-wont-replace-software-engineers) 23-Mar-2023 'The iPhone Moment of AI' Nvidia to Rent Out Supercomputers Behind ChatGPT to Businesses for $37,000 a Month Bill Gates calls AI revolutionary, says it can reduce some of the world’s worst inequities AI pics of Donald Trump's arrest by 'cop' Joe Biden go viral. Will we no longer be able to tell what’s real vs what’s fake?” - Eluna AI New research shows we can only accurately identify AI writers about 50% of the time. (https://hai.stanford.edu/news/was-written-human-or-ai-tsu) FauxPilot - an open-source GitHub Copilot server(https://github.com/fauxpilot/fauxpilot) Flower , an open-source framework for training AI on distributed data. We move the model to the data instead of moving the data to the model. (https://flower.dev/) OpenAI-Integrated Microsoft Bing Outperforms Google in Page Visits (https://www.gadgets360.com/internet/news/openai-integrated-microsoft-bing-outperforms-google-page-visits-growth-3885069) GitHub Copilot X: GitHub Copilot is evolving to bring chat and voice interfaces, support pull requests, answer questions on docs, and adopt OpenAI’s GPT-4 for a more personalized developer experience. (https://github.blog/2023-03-22-github-copilot-x-the-ai-powered-developer-experience/) Moonshine – open-source, pretrained ML models for satellite (https://github.com/moonshinelabs-ai/moonshine) Mozilla.ai: A startup — and a community — that will build a trustworthy and independent open-source AI ecosystem. Mozilla.ai’s initial focus? Tools that make generative AI safer and more transparent. And, people-centric recommendation systems that don’t misinform or undermine our well-being. (https://blog.mozilla.org/en/mozilla/introducing-mozilla-ai-investing-in-trustworthy-ai/) OpenAI’s policies hinder reproducible research on language models (https://aisnakeoil.substack.com/p/openais-policies-hinder-reproducible) 24-Mar-2023 Adobe has added AI features to Photoshop and Illustrator, while Nvidia has unveiled ‘Picasso’ AI image generation service. ChatGPT-owner OpenAI fixes 'significant issue' exposing user chat titles.A bug in an open-source library caused ChatGPT to leak user conversation titles. Graphic design platform Canva introduces new generative AI tools Gmail for Android, Google Messages to Soon Get Features for AI-Generated Texts Apple: Transformer architecture optimized for Apple Silicon (https://github.com/apple/ml-ane-transformers) ChatGPT plugins, join waitlist (https://openai.com/blog/chatgpt-plugins) Microsoft's paper on OpenAI's GPT-4 had hidden information (https://twitter.com/DV2559106965076/status/1638769434763608064) how to use LoRA to fine-tune LLaMA using Alpaca training data (https://replicate.com/blog/fine-tune-alpaca-with-lora) Helicone: one-line integration logs the prompts, completions, latencies, and costs of your OpenAI requests (https://github.com/Helicone/helicone) RWKV is an RNN with Transformer-level LLM performance, which can also be directly trained like a GPT transformer (parallelizable). (https://github.com/BlinkDL/RWKV-LM) open-source retrieval plugin The open-source retrieval plugin enables ChatGPT to access personal or organizational information sources (with permission). It allows users to obtain the most relevant document snippets from their data sources, such as files, notes, emails or public documentation, by asking questions or expressing needs in natural language. Security considerations The retrieval plugin allows ChatGPT to search a vector database of content, and add the best results into the ChatGPT session. This means it doesn’t have any external effects, and the main risk is data authorization and privacy. Developers should only add content into their retrieval plugin that they are authorized to use and can share in users’ ChatGPT sessions. https://github.com/openai/chatgpt-retrieval-plugin 27-Mar-2023 Autodoc: Toolkit for auto-generating codebase documentation using LLMs (https://github.com/context-labs/autodoc) March 20 ChatGPT outage: Here’s what happened (https://openai.com/blog/march-20-chatgpt-outage) Facebook is going after LLaMA repos with DMCA's (https://twitter.com/theshawwn/status/1638925249709240322) ChatGPT + Wolfram is INSANE! (https://old.reddit.com/r/ChatGPT/comments/1205omc/chatgpt\_wolfram\_is\_insane/) Reproducing the Stanford Alpaca results using low-rank adaptation (LoRA) (https://github.com/chris-alexiuk/alpaca-lora) GOAT, a decentralized way to publish and download AI models.Powered by BitTorrent and Bitcoin.(https://ipfs.io/ipfs/QmYyucgBQVfs9JXZ2MtmkGPAhgUjNgyGE6rcJT1KybQHhp/index.html) Dolly from databricks (https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html) AI powered Developer Tools 2.0. https://www.sequoiacap.com/article/ai-powered-developer-tools/ Turn your designs into production-ready front-end code for mobile apps and the web (https://www.locofy.ai/) Using ChatGPT Plugins with LLaMA (https://blog.lastmileai.dev/using-openais-retrieval-plugin-with-llama-d2e0b6732f14) 28-Mar-2023 Bing AI now allows 20 prompts per session and can make images for you ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks (https://arxiv.org/abs/2303.15056) ChatGPT or Grammarly? Evaluating ChatGPT on Grammatical Error Correction Benchmark (https://arxiv.org/abs/2303.13648) AI-controlled Linux Containers (https://github.com/fafrd/aquarium) Microsoft reportedly orders AI chatbot rivals to stop using Bing’s search data (https://www.theverge.com/2023/3/25/23656336/microsoft-chatbot-rivals-stop-using-bing-search-index) 29-Mar-2023 Text2Video-Zero Code and Weights Released by Picsart AI Research (12G VRAM).(https://github.com/Picsart-AI-Research/Text2Video-Zero) Pause Giant AI Experiments: An Open Letter. Huggingface's SF Open-Source AI Meetup officially has 2000 people registered. Cerebras open sources seven GPT-3 models from 111 million to 13 billion parameters. Trained using the Chinchilla formula, these models set new benchmarks for accuracy and compute efficiency.(https://www.cerebras.net/blog/cerebras-gpt-a-family-of-open-compute-efficient-large-language-models/) Independent implementation of LLaMA that is fully open source under the Apache 2.0 license (https://github.com/Lightning-AI/lit-llama) Bootstrap knowledge of LLMs (https://gist.github.com/rain-1/eebd5e5eb2784feecf450324e3341c8d) OPENFLAMINGO: AN OPEN-SOURCE FRAMEWORK FOR TRAINING VISION-LANGUAGE MODELS WITH IN-CONTEXT LEARNING (https://laion.ai/blog/open-flamingo/) gpt4all: a chatbot trained on a massive collection of clean assistant data including code, stories and dialogue (https://github.com/nomic-ai/gpt4all) 30-Mar-2022 Microsoft Security Copilot is a new GPT-4 AI assistant for cybersecurity (https://www.theverge.com/2023/3/28/23659711/microsoft-security-copilot-gpt-4-ai-tool-features) UK details ‘pro-innovation’ approach to AI regulation (https://www.artificialintelligence-news.com/2023/03/29/uk-details-pro-innovation-approach-ai-regulation/) Employees Are Feeding Sensitive Biz Data to ChatGPT, Raising Security Fears (https://www.darkreading.com/risk/employees-feeding-sensitive-business-data-chatgpt-raising-security-fears) In the Age of AI, Don't Let Your Skills Atrophy (https://www.cyberdemon.org/2023/03/29/age-of-ai-skill-atrophy.html) Now ChatGPT is being (mis)used to do #PeerReview (https://mstdn.science/@ukrio/110100752908161183) Bing Chat now has Ads! (https://twitter.com/debarghya\_das/status/1640892791923572737) Cerebras-GPT vs LLaMA AI Model Comparison (https://www.lunasec.io/docs/blog/cerebras-gpt-vs-llama-ai-model-comparison/) Arthur C. Clarke about the future of AI. — 21 September 1964 (https://twitter.com/Rainmaker1973/status/1640016339011076097) ColossalChat: An Open-Source Solution for Cloning ChatGPT With a Complete RLHF Pipeline (https://medium.com/@yangyou\_berkeley/colossalchat-an-open-source-solution-for-cloning-chatgpt-with-a-complete-rlhf-pipeline-5edf08fb538b) Create and Embed Custom AI Assistants with Libraria (https://libraria.dev/) 31-Mar-2023 Deranged New AI Has No Guardrails Whatsoever, Proudly Praises Hitler (https://futurism.com/deranged-ai-no-guardrails) Midjourney Kills Free AI Image Generator Access After Explosion of Deep Fakes (https://decrypt.co/124972/midjourney-free-ai-image-generation-stopped-over-deepfakes) Judge asks ChatGPT to decide bail in murder trial (https://nypost.com/2023/03/29/judge-asks-chatgpt-for-decision-in-murder-trial/) Should you use OpenAI's embeddings? Probably not, and here's why. (https://iamnotarobot.substack.com/p/should-you-use-openais-embeddings) Visual Studio Code and GitHub Copilot (https://code.visualstudio.com/blogs/2023/03/30/vscode-copilot) Llama Hub (https://llamahub.ai/) Finetuning LLMs on a Single GPU Using Gradient Accumulation (https://lightning.ai/pages/blog/gradient-accumulation/) Open source ETL framework for retrieval augmented generation (RAG). Sync data from your SaaS tools to a vector store, where they can be easily queried by GPT apps (https://github.com/ai-sidekick/sidekick) HALTT4LLM - Hallucination Trivia Test for Large Language Models (https://github.com/manyoso/haltt4llm) Vicuna: An Open-Source Chatbot Impressing GPT-4 with 90%* ChatGPT Quality (https://vicuna.lmsys.org/) Iterate.ai Brings Generative AI Capabilities to Interplay, the Low-Code Platform Accelerating Customers’ Digital Innovation (https://www.indianweb2.com/2023/03/iterateai-brings-generative-ai.html) RFdiffusion is an open source method for structure generation, with or without conditional information (a motif, target etc). (https://github.com/RosettaCommons/RFdiffusion) Google denies training Bard on ChatGPT chats from ShareGPT
-
[D] An ELI5 explanation for LoRA - Low-Rank Adaptation.
Repos like https://github.com/tloen/alpaca-lora and https://github.com/Lightning-AI/lit-llama use LoRA as a method to fine-tune LLaMA models.
- Together’s $20M seed funding to build open-source AI and cloud platform
-
If you had ten minutes alone in a store where everything was going to be paid for what would you choose?
A high-end computer store where I could grab as many NVIDIA AI GPU's as possible to start running my own AI LLM at home.
What are some alternatives?
server - The Triton Inference Server provides an optimized cloud and edge inferencing solution.
StableLM - StableLM: Stability AI Language Models
MNN - MNN is a blazing fast, lightweight deep learning framework, battle-tested by business-critical use cases in Alibaba
fauxpilot - FauxPilot - an open-source alternative to GitHub Copilot server
flashlight - A C++ standalone library for machine learning
godot-dodo - Finetuning large language models for GDScript generation.
XLA.jl - Julia on TPUs
flake - A Nix flake for many AI projects
oneflow - OneFlow is a deep learning framework designed to be user-friendly, scalable and efficient.
pair - REPL environment for GPT pair programming
glow - Compiler for Neural Network hardware accelerators
sidekick - Universal APIs for unstructured data. Sync documents from SaaS tools to a SQL or vector database, where they can be easily queried by AI applications [Moved to: https://github.com/psychic-api/psychic]