open weights

llms-keep-leaping-with-llama-3,-meta’s-newest-open-weights-ai-model

LLMs keep leaping with Llama 3, Meta’s newest open-weights AI model

AI, AI assistants, Biz & IT, chatgpt, chatgtp, copilot, Google Gemini, LLaMA, Llama 2, Llama 3, machine learning, Meta, Meta AI Assistant, microsoft, Microsoft Copilot, Open Source, open weights, openai / /

computer-powered word generator —

Zuckerberg says new AI model “was still learning” when Meta stopped training.

–

A group of pink llamas on a pixelated background.

On Thursday, Meta unveiled early versions of its Llama 3 open-weights AI model that can be used to power text composition, code generation, or chatbots. It also announced that its Meta AI Assistant is now available on a website and is going to be integrated into its major social media apps, intensifying the company’s efforts to position its products against other AI assistants like OpenAI’s ChatGPT, Microsoft’s Copilot, and Google’s Gemini.

Like its predecessor, Llama 2, Llama 3 is notable for being a freely available, open-weights large language model (LLM) provided by a major AI company. Llama 3 technically does not quality as “open source” because that term has a specific meaning in software (as we have mentioned in other coverage), and the industry has not yet settled on terminology for AI model releases that ship either code or weights with restrictions (you can read Llama 3’s license here) or that ship without providing training data. We typically call these releases “open weights” instead.

At the moment, Llama 3 is available in two parameter sizes: 8 billion (8B) and 70 billion (70B), both of which are available as free downloads through Meta’s website with a sign-up. Llama 3 comes in two versions: pre-trained (basically the raw, next-token-prediction model) and instruction-tuned (fine-tuned to follow user instructions). Each has a 8,192 token context limit.

A screenshot of the Meta AI Assistant website on April 18, 2024.

Enlarge / A screenshot of the Meta AI Assistant website on April 18, 2024.

Benj Edwards

Meta trained both models on two custom-built, 24,000-GPU clusters. In a podcast interview with Dwarkesh Patel, Meta CEO Mark Zuckerberg said that the company trained the 70B model with around 15 trillion tokens of data. Throughout the process, the model never reached “saturation” (that is, it never hit a wall in terms of capability increases). Eventually, Meta pulled the plug and moved on to training other models.

“I guess our prediction going in was that it was going to asymptote more, but even by the end it was still leaning. We probably could have fed it more tokens, and it would have gotten somewhat better,” Zuckerberg said on the podcast.

Meta also announced that it is currently training a 400B parameter version of Llama 3, which some experts like Nvidia’s Jim Fan think may perform in the same league as GPT-4 Turbo, Claude 3 Opus, and Gemini Ultra on benchmarks like MMLU, GPQA, HumanEval, and MATH.

Speaking of benchmarks, we have devoted many words in the past to explaining how frustratingly imprecise benchmarks can be when applied to large language models due to issues like training contamination (that is, including benchmark test questions in the training dataset), cherry-picking on the part of vendors, and an inability to capture AI’s general usefulness in an interactive session with chat-tuned models.

But, as expected, Meta provided some benchmarks for Llama 3 that list results from MMLU (undergraduate level knowledge), GSM-8K (grade-school math), HumanEval (coding), GPQA (graduate-level questions), and MATH (math word problems). These show the 8B model performing well compared to open-weights models like Google’s Gemma 7B and Mistral 7B Instruct, and the 70B model also held its own against Gemini Pro 1.5 and Claude 3 Sonnet.

A chart of instruction-tuned Llama 3 8B and 70B benchmarks provided by Meta.

Enlarge / A chart of instruction-tuned Llama 3 8B and 70B benchmarks provided by Meta.

Meta says that the Llama 3 model has been enhanced with capabilities to understand coding (like Llama 2) and, for the first time, has been trained with both images and text—though it currently outputs only text. According to Reuters, Meta Chief Product Officer Chris Cox noted in an interview that more complex processing abilities (like executing multi-step plans) are expected in future updates to Llama 3, which will also support multimodal outputs—that is, both text and images.

Meta plans to host the Llama 3 models on a range of cloud platforms, making them accessible through AWS, Databricks, Google Cloud, and other major providers.

Also on Thursday, Meta announced that Llama 3 will become the new basis of the Meta AI virtual assistant, which the company first announced in September. The assistant will appear prominently in search features for Facebook, Instagram, WhatsApp, Messenger, and the aforementioned dedicated website that features a design similar to ChatGPT, including the ability to generate images in the same interface. The company also announced a partnership with Google to integrate real-time search results into the Meta AI assistant, adding to an existing partnership with Microsoft’s Bing.

LLMs keep leaping with Llama 3, Meta’s newest open-weights AI model Read More »

stability-announces-stable-diffusion-3,-a-next-gen-ai-image-generator

Stability announces Stable Diffusion 3, a next-gen AI image generator

AI, AI image generators, Biz & IT, dall-e, DALL-E 3, Deepfakes, Emad Mostaque, image synthesis, machine learning, open weights, openai, SDXL, source available, Stability AI, Stable Diffusion, Stable Diffusion 3, Stable Diffusion XL / /

Pics and it didn’t happen —

SD3 may bring DALL-E-like prompt fidelity to an open-weights image-synthesis model.

–

Stable Diffusion 3 generation with the prompt: studio photograph closeup of a chameleon over a black background.

Enlarge / Stable Diffusion 3 generation with the prompt: studio photograph closeup of a chameleon over a black background.

On Thursday, Stability AI announced Stable Diffusion 3, an open-weights next-generation image-synthesis model. It follows its predecessors by reportedly generating detailed, multi-subject images with improved quality and accuracy in text generation. The brief announcement was not accompanied by a public demo, but Stability is opening up a waitlist today for those who would like to try it.

Stability says that its Stable Diffusion 3 family of models (which takes text descriptions called “prompts” and turns them into matching images) range in size from 800 million to 8 billion parameters. The size range accommodates allowing different versions of the model to run locally on a variety of devices—from smartphones to servers. Parameter size roughly corresponds to model capability in terms of how much detail it can generate. Larger models also require more VRAM on GPU accelerators to run.

Since 2022, we’ve seen Stability launch a progression of AI image-generation models: Stable Diffusion 1.4, 1.5, 2.0, 2.1, XL, XL Turbo, and now 3. Stability has made a name for itself as providing a more open alternative to proprietary image-synthesis models like OpenAI’s DALL-E 3, though not without controversy due to the use of copyrighted training data, bias, and the potential for abuse. (This has led to lawsuits that are unresolved.) Stable Diffusion models have been open-weights and source-available, which means the models can be run locally and fine-tuned to change their outputs.

  • Stable Diffusion 3 generation with the prompt: Epic anime artwork of a wizard atop a mountain at night casting a cosmic spell into the dark sky that says “Stable Diffusion 3” made out of colorful energy.

  • An AI-generated image of a grandma wearing a “Go big or go home sweatshirt” generated by Stable Diffusion 3.

  • Stable Diffusion 3 generation with the prompt: Three transparent glass bottles on a wooden table. The one on the left has red liquid and the number 1. The one in the middle has blue liquid and the number 2. The one on the right has green liquid and the number 3.

  • An AI-generated image created by Stable Diffusion 3.

  • Stable Diffusion 3 generation with the prompt: A horse balancing on top of a colorful ball in a field with green grass and a mountain in the background.

  • Stable Diffusion 3 generation with the prompt: Moody still life of assorted pumpkins.

  • Stable Diffusion 3 generation with the prompt: a painting of an astronaut riding a pig wearing a tutu holding a pink umbrella, on the ground next to the pig is a robin bird wearing a top hat, in the corner are the words “stable diffusion.”

  • Stable Diffusion 3 generation with the prompt: Resting on the kitchen table is an embroidered cloth with the text ‘good night’ and an embroidered baby tiger. Next to the cloth there is a lit candle. The lighting is dim and dramatic.

  • Stable Diffusion 3 generation with the prompt: Photo of an 90’s desktop computer on a work desk, on the computer screen it says “welcome”. On the wall in the background we see beautiful graffiti with the text “SD3” very large on the wall.

As far as tech improvements are concerned, Stability CEO Emad Mostaque wrote on X, “This uses a new type of diffusion transformer (similar to Sora) combined with flow matching and other improvements. This takes advantage of transformer improvements & can not only scale further but accept multimodal inputs.”

Like Mostaque said, the Stable Diffusion 3 family uses diffusion transformer architecture, which is a new way of creating images with AI that swaps out the usual image-building blocks (such as U-Net architecture) for a system that works on small pieces of the picture. The method was inspired by transformers, which are good at handling patterns and sequences. This approach not only scales up efficiently but also reportedly produces higher-quality images.

Stable Diffusion 3 also utilizes “flow matching,” which is a technique for creating AI models that can generate images by learning how to transition from random noise to a structured image smoothly. It does this without needing to simulate every step of the process, instead focusing on the overall direction or flow that the image creation should follow.

A comparison of outputs between OpenAI's DALL-E 3 and Stable Diffusion 3 with the prompt,

Enlarge / A comparison of outputs between OpenAI’s DALL-E 3 and Stable Diffusion 3 with the prompt, “Night photo of a sports car with the text “SD3″ on the side, the car is on a race track at high speed, a huge road sign with the text ‘faster.'”

We do not have access to Stable Diffusion 3 (SD3), but from samples we found posted on Stability’s website and associated social media accounts, the generations appear roughly comparable to other state-of-the-art image-synthesis models at the moment, including the aforementioned DALL-E 3, Adobe Firefly, Imagine with Meta AI, Midjourney, and Google Imagen.

SD3 appears to handle text generation very well in the examples provided by others, which are potentially cherry-picked. Text generation was a particular weakness of earlier image-synthesis models, so an improvement to that capability in a free model is a big deal. Also, prompt fidelity (how closely it follows descriptions in prompts) seems to be similar to DALL-E 3, but we haven’t tested that ourselves yet.

While Stable Diffusion 3 isn’t widely available, Stability says that once testing is complete, its weights will be free to download and run locally. “This preview phase, as with previous models,” Stability writes, “is crucial for gathering insights to improve its performance and safety ahead of an open release.”

Stability has been experimenting with a variety of image-synthesis architectures recently. Aside from SDXL and SDXL Turbo, just last week, the company announced Stable Cascade, which uses a three-stage process for text-to-image synthesis.

Listing image by Emad Mostaque (Stability AI)

Stability announces Stable Diffusion 3, a next-gen AI image generator Read More »