The AMD Instinct MI50 (~$110) is surprisingly fast for inference Quantized models.

This runs a Llama 3.1 8B Q8 with Llama.cpp
DevQuasar/Mi50

A little blogpost about the HW
http://devquasar.com/uncategorized/amd-radeon-instinct-mi50-cheap-inference/

reacted to m-ric's post with 🚀 15 days ago

Post

2425

💥 𝗚𝗼𝗼𝗴𝗹𝗲 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝘀 𝗚𝗲𝗺𝗶𝗻𝗶 𝟮.𝟬, 𝘀𝘁𝗮𝗿𝘁𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗮 𝗙𝗹𝗮𝘀𝗵 𝗺𝗼𝗱𝗲𝗹 𝘁𝗵𝗮𝘁 𝘀𝘁𝗲𝗮𝗺𝗿𝗼𝗹𝗹𝘀 𝗚𝗣𝗧-𝟰𝗼 𝗮𝗻𝗱 𝗖𝗹𝗮𝘂𝗱𝗲-𝟯.𝟲 𝗦𝗼𝗻𝗻𝗲𝘁! And they start a huge effort on agentic capabilities.

🚀 The performance improvements are crazy for such a fast model:
‣ Gemini 2.0 Flash outperforms the previous 1.5 Pro model at twice the speed
‣ Now supports both input AND output of images, video, audio and text
‣ Can natively use tools like Google Search and execute code

➡️ If the price is on par with previous Flash iteration ($0.30 / M tokens, to compare with GPT-4o's $1.25) the competition will have a big problem with this 4x cheaper model that gets better benchmarks 🤯

🤖 What about the agentic capabilities?

‣ Project Astra: A universal AI assistant that can use Google Search, Lens and Maps
‣ Project Mariner: A Chrome extension that can complete complex web tasks (83.5% success rate on WebVoyager benchmark, this is really impressive!)
‣ Jules: An AI coding agent that integrates with GitHub workflows

I'll be eagerly awaiting further news from Google!

Read their blogpost here 👉 https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/

reacted to merve's post with ❤️ 17 days ago

Post

5514

This week in open-source AI was insane 🤠 A small recap🕺🏻 merve/dec-6-releases-67545caebe9fc4776faac0a3

Multimodal 🖼️
> Google shipped a PaliGemma 2, new iteration of PaliGemma with more sizes: 3B, 10B and 28B, with pre-trained and captioning variants 👏
> OpenGVLab released InternVL2, seven new vision LMs in different sizes, with sota checkpoint with MIT license ✨
> Qwen team at Alibaba released the base models of Qwen2VL models with 2B, 7B and 72B ckpts

LLMs 💬
> Meta released a new iteration of Llama 70B, Llama3.2-70B trained further
> EuroLLM-9B-Instruct is a new multilingual LLM for European languages with Apache 2.0 license 🔥
> Dataset: CohereForAI released GlobalMMLU, multilingual version of MMLU with 42 languages with Apache 2.0 license
> Dataset: QwQ-LongCoT-130K is a new dataset to train reasoning models
> Dataset: FineWeb2 just landed with multilinguality update! 🔥 nearly 8TB pretraining data in many languages!

Image/Video Generation 🖼️
> Tencent released HunyuanVideo, a new photorealistic video generation model
> OminiControl is a new editing/control framework for image generation models like Flux

Audio 🔊
> Indic-Parler-TTS is a new text2speech model made by community

reacted to Xenova's post with 🚀 17 days ago

Post

2589

Introducing TTS WebGPU: The first ever text-to-speech web app built with WebGPU acceleration! 🔥 High-quality and natural speech generation that runs 100% locally in your browser, powered by OuteTTS and Transformers.js. 🤗 Try it out yourself!

Demo: webml-community/text-to-speech-webgpu
Source code: https://github.com/huggingface/transformers.js-examples/tree/main/text-to-speech-webgpu
Model: onnx-community/OuteTTS-0.2-500M (ONNX), OuteAI/OuteTTS-0.2-500M (PyTorch)

upvoted a collection 19 days ago

Recommended small models

Collection

This is everything recent smaller than ~25B parameters that are high quality/reputable • 19 items • Updated 26 days ago • 24

liked a model 22 days ago

showlab/ShowUI-2B

Updated 22 days ago • 14.8k • 210

upvoted a collection 22 days ago

🤖 Agents

Collection

21 items • Updated 16 days ago • 57

reacted to m-ric's post with 👀 22 days ago

Post

1468

𝗦𝗵𝗼𝘄𝗨𝗜: 𝗮 𝘀𝗺𝗮𝗹𝗹 𝗲𝗻𝗱-𝘁𝗼-𝗲𝗻𝗱 𝗮𝗴𝗲𝗻𝘁 𝘁𝗵𝗮𝘁 𝗰𝗮𝗻 𝗻𝗮𝘃𝗶𝗴𝗮𝘁𝗲 𝗮𝗻𝘆 𝗨𝗜 𝗮𝗻𝗱 𝗼𝘂𝘁𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝘀 𝗺𝘂𝗰𝗵 𝗯𝗶𝗴𝗴𝗲𝗿 𝘀𝘆𝘀𝘁𝗲𝗺𝘀! 📲

A team from NUS and Microsoft just released an agent that can act on any UI (Desktop, Android, Web) without needing additional text information. It works extremely well : they applied their method on a tiny Qwen2-VL-2B, and they managed to beat methods that use either much more powerful vision models (like GPT-4V) without using any additional info (e.g. leveraging the DOM of a webpage) like previous methods did ! 👏👏

They started from the idea that most existing methods rely heavily on text, which makes them less generalizable, while letting aside rich UI structure that user actually rely on when navigating this interfaces.

⚙️ They put several good ideas to work:

💡 Simplify screenshots to the max:
They prune a lot the heavy visual content of UI screenshots, by removing cloned image patches (like any vast patch of the same color will be reduced to a small patch, while maintaining positional embeddings), then group patches from the same GUI elements together to simplify even further

💡 Build a truly generalist dataset:
To train a general UI agent, you need trajectories from each possible UI, and express them in a common language. Authors merge datasets like OmniAct for Desktop, Mind2Web for websites, AMEX for Android trajectories to create a high-quality and diverse dataset.

➡️ Nice results ensued:
They fine-tune a tiny Qwen-2-VL-2B on their method, and it reaches SOTA on several task (element identification, web navigation), even beating methods that either use additional info from the DOM or use much bigger VLMS like GPT-4v! 🏆

And performance could certainly jump with a slightly bigger vision model. Let's hope the community builds this soon! 🚀

Paper added to my "Agents" collection 👉 m-ric/agents-65ba776fbd9e29f771c07d4e

reacted to Xenova's post with 🚀 28 days ago

Post

3928

We just released Transformers.js v3.1 and you're not going to believe what's now possible in the browser w/ WebGPU! 🤯 Let's take a look:
🔀 Janus from Deepseek for unified multimodal understanding and generation (Text-to-Image and Image-Text-to-Text)
👁️ Qwen2-VL from Qwen for dynamic-resolution image understanding
🔢 JinaCLIP from Jina AI for general-purpose multilingual multimodal embeddings
🌋 LLaVA-OneVision from ByteDance for Image-Text-to-Text generation
🤸‍♀️ ViTPose for pose estimation
📄 MGP-STR for optical character recognition (OCR)
📈 PatchTST & PatchTSMixer for time series forecasting

That's right, everything running 100% locally in your browser (no data sent to a server)! 🔥 Huge for privacy!

Check out the release notes for more information. 👇
https://github.com/huggingface/transformers.js/releases/tag/3.1.0

Demo link (+ source code): webml-community/Janus-1.3B-WebGPU

reacted to PLB's post with 🚀 about 1 month ago

Post

1863

⚠️ People selling AI chatbots for websites hate us.
Add an open source chat assistant on your website in 5 minutes: https://github.com/phospho-app/ai-chat-bubble

How does it work ?
- You give an URL
- The AI assistant crawls the website content and embed it
- Add it to your frontend in one line of code
- People on your website can ask the assistant questions

Powered by BAAI/bge-small-en-v1.5 and Mistral AI

5 replies

reacted to fdaudens's post with 👍 about 1 month ago

Post

1844

Been reading about the "bigger models = better AI" narrative getting pushed back today.

@thomwolf tackled this head on at Web Summit and highlighted how important small models are (and why closed-source companies haven't pushed for this 😬). They're crushing it: today's 1B parameter models outperform last year's 10B models.

Fascinating to hear him talk about the secret sauce behind this approach.

upvoted a collection about 1 month ago

Qwen2.5

Collection

Qwen2.5 language models, including pretrained and instruction-tuned models of 7 sizes, including 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B. • 45 items • Updated 29 days ago • 445

liked a model about 2 months ago

Kortix/FastApply-1.5B-v1.0

Text Generation • Updated Oct 27 • 5.03k • 17

reacted to fdaudens's post with 👍 about 2 months ago

Post

2291

🔍 NYT leveraged AI to investigate election interference by analyzing 400+ hours of recorded meetings - that's 5M words of data!

AI spotted patterns, humans verified facts. Every AI-flagged quote was manually verified against source recordings. Really appreciate that they published their full methodology - transparency matters when using AI in journalism.

A perfect blend of tech & journalism.

The future of journalism isn't robots replacing reporters - it's AI helping humans process massive datasets more efficiently. Sometimes the most powerful tech solutions are the least flashy ones.

Read the article: https://www.nytimes.com/interactive/2024/10/28/us/politics/inside-the-movement-behind-trumps-election-lies.html?unlocked_article_code=1.Vk4.ucv9.dbHVquTQaf0G&smid=nytcore-ios-share

reacted to m-ric's post with 👀 2 months ago

Post

864

𝗛𝗼𝘄 𝘁𝗼 𝗿𝗲-𝗿𝗮𝗻𝗸 𝘆𝗼𝘂𝗿 𝘀𝗻𝗶𝗽𝗽𝗲𝘁𝘀 𝗶𝗻 𝗥𝗔𝗚 ⇒ ColBERT, Rerankers, Cross-Encoders

Let’s say you’re doing RAG, and in an effort to improve performance, you try to rerank a few possible source snippets by their relevancy to a query.

How can you score similarity between your query and any source document? 🤔 📄 ↔️ 📑

𝟭. 𝗝𝘂𝘀𝘁 𝘂𝘀𝗲 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 : 𝗡𝗼-𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻 🏎️

This means that you encode each token from both the query and the doc as separate vectors, then average the tokens of each separately to get in total 2 vectors, then you compute similarity via cosine or something.
➡️ Notable examples: Check the top of the MTEB leaderboard!

𝟮. 𝗟𝗮𝘁𝗲-𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻: 𝘁𝗵𝗶𝘀 𝗶𝘀 𝗖𝗼𝗹𝗕𝗘𝗥𝗧 🏃

These encode each token from both query and doc as separate vectors as before, but compare all together without previously averaging them and losing information.

This is more accurate than no-interaction but also slower because you have to compare n*m vectors instead of 2. At least you can store documents in memory. And ColBERT has some optimisations like pooling to be faster.

➡️ Notable examples: ColBERTv2, mxbai-colbert-large-v1, jina-colbert-v2

𝟯. 𝗘𝗮𝗿𝗹𝘆 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻: 𝗖𝗿𝗼𝘀𝘀-𝗲𝗻𝗰𝗼𝗱𝗲𝗿𝘀 🏋️

Basically you run the concatenated query + document in a model to get a final score.

This is the most accurate, but also the slowest since it gets really long when you have many docs to rerank! And you cannot pre-store embeddings.

➡️ Notable examples: MixedBread or Jina AI rerankers!

🚀 So what you choose is a trade-off between speed and accuracy: I think ColBERT is often a really good choice!

Based on this great post by Jina AI 👉 https://jina.ai/news/what-is-colbert-and-late-interaction-and-why-they-matter

liked a model 3 months ago

openai/whisper-large-v3-turbo

Automatic Speech Recognition • Updated Oct 4 • 2.54M • • 1.66k

reacted to m-ric's post with 👀 3 months ago

Post

367

Anthropic just released a chunk improvement technique that vastly improves RAG performance! 🔥

Crash reminder: Retrieval Augmented Generation (RAG) is a widely-used technique for improving your LLM chatbot's answers to user questions.

It goes like this: instead of generating an LLM answer straight away, it just adds a previous step called Retrieval, that retrieves relevant documents from your knowledge base through semantic search, and just appends the top K documents to the prompt. ➡️ As a result, the LLM answer is grounded in context.

⛔️ The difficulty with this retrieval step is that when you split your documents into chunks that will be retrieved, you lose context. So importance chunks could be missed.

💡 Anthropic's just released blog post shows that you can add some context to each chunk, with one LLM call. Then you embed the original chunk + a bit of added context, so that the embedding is much more representative of the document in its context!

🤔 Isn't that crazy expensive? Well it would have been before, but not so much anymore with their new Prompt caching feature that makes duplicating thousands of requests with the same prompt much less expensive. They give an indicative price tag of only $1.02 per million chunks processed!

✅ And this vastly improves performance on their benchmark!

Read their blog post 👉 https://www.anthropic.com/news/contextual-retrieval