正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

This is a new diffusion-based language model from Inception. I don't know how well it works compared to an autoregressive LM of a similar size, but the inference process looks badass:

BURKOV

52,262 subscribers

58,434 次观看 • 1 年前 •via X (Twitter)

科学技术新闻政治教育

Anya Rossi• Live Now

Private livecam show

11 条评论

wabalabadubd 的头像

wabalabadubd1 年前

@InceptionAILabs Finally something which is not just one upping the benchmarks.

Malarkey Roofing Products 的头像

Malarkey Roofing Products1 年前

A little more thought, a little more effort, a little more care. If there’s a better, more sustainable way to make a shingle, we intend to find it. #AnythingButStandard #Innovate #EcoFriendlyRoofing #Greener #SustainableInnovation #SustainableRoofing #SustainableShingles @Holcim

TheFullStackSnack 的头像

TheFullStackSnack1 年前

@InceptionAILabs The fact that approach like "start with a random set of characters and tweak until it works" is working for coding blow my mind.

Dr. Novo 🍌 的头像

Dr. Novo 🍌1 年前

I’m beyond excited by their work! Been working on adapting a GAN model to work with discrete data (since 2023) but it’s been a nightmare! It’s genius actually to use diffusion models to capitalise on noise to boost the quality and speed of outputs in LLMs! I’m just dying to know how they solved the discrete/continuous duality in practice!

Qusai Ismael 的头像

Qusai Ismael1 年前

@InceptionAILabs Look at the way it generates the answer, this would be a large shift from the old way. imagine adding thinking to it too.

John Januszczak 的头像

John Januszczak1 年前

@InceptionAILabs I asked it to create a react component. It queued the request (probably because if everyone on X trying it), but when it started it was lightning fast.

cargo 的头像

cargo1 年前

@InceptionAILabs what is the advantage or disadvantage of using diffusion here?

Tobias Holgersen 的头像

Tobias Holgersen1 年前

@InceptionAILabs They should make it even more like the matrix 💪

Bhaktavaschal Samal 的头像

Bhaktavaschal Samal1 年前

@InceptionAILabs actually, worked on this for a few days, gave up in the end

Midas 霊 👑 (kween/acc) 的头像

Midas 霊 👑 (kween/acc)1 年前

@InceptionAILabs don't miss this one too, also launched on the same day seems like something's in the air with these dllms

Rogerio Chaves 的头像

Rogerio Chaves1 年前

@InceptionAILabs Yeah, I don’t get it, why do you need to enable the “diffusion effects”? Seems like a frontend gimmick otherwise it’s just like other LLMs?

相关视频

MinerU-Diffusion A 2.5B diffusion-based OCR model that replaces slow autoregressive decoding with parallel block-wise diffusion, achieving up to 3.2x faster inference while improving robustness on complex documents with tables, formulas, and layouts.

MinerU-Diffusion A 2.5B diffusion-based OCR model that replaces slow autoregressive decoding with parallel block-wise diffusion, achieving up to 3.2x faster inference while improving robustness on complex documents with tables, formulas, and layouts.

DailyPapers

15,304 次观看 • 2 个月前

Results from a failed attempt to train a latent diffusion model on the results of my gabor image fitting process. The model doesn't do what it was supposed to, but I don't really mind what it's doing.

Results from a failed attempt to train a latent diffusion model on the results of my gabor image fitting process. The model doesn't do what it was supposed to, but I don't really mind what it's doing.

Andrew 😵‍💫

55,609 次观看 • 1 个月前

Meta announces SceneScript Reconstructing Scenes With An Autoregressive Structured Language Model We introduce SceneScript, a method that directly produces full scene models as a sequence of structured language commands using an autoregressive, token-based approach. Our

Meta announces SceneScript Reconstructing Scenes With An Autoregressive Structured Language Model We introduce SceneScript, a method that directly produces full scene models as a sequence of structured language commands using an autoregressive, token-based approach. Our

AK

46,090 次观看 • 2 年前

Most don't know (1) how easy it is to invert embedding vectors back into sentences, (2) this is a perfect task text diffusion models. Here's a 78M parameter model and live demo that recovers 80% of tokens from Qwen3-Embedding and EmbeddingGemma vectors. Works even on multilingual input.

Most don't know (1) how easy it is to invert embedding vectors back into sentences, (2) this is a perfect task text diffusion models. Here's a 78M parameter model and live demo that recovers 80% of tokens from Qwen3-Embedding and EmbeddingGemma vectors. Works even on multilingual input.

Jina AI

12,805 次观看 • 4 个月前

I don't speak the language but I know "LOOK AT THE F'ING SIZE OF IT!" when I hear it

I don't speak the language but I know "LOOK AT THE F'ING SIZE OF IT!" when I hear it

Ribz of Tik Tok

118,740 次观看 • 4 个月前

Block Diffusion Interpolating Between Autoregressive and Diffusion Language Models

Block Diffusion Interpolating Between Autoregressive and Diffusion Language Models

AK

160,553 次观看 • 1 年前

Diffusion language models are SO FAST!! A new startup, Inception Labs, has released Mercury Coder, "the first commercial-scale diffusion large language model" It's 5-10x faster than current gen LLMs, providing high-quality responses at low costs. And you can try it now!

Diffusion language models are SO FAST!! A new startup, Inception Labs, has released Mercury Coder, "the first commercial-scale diffusion large language model" It's 5-10x faster than current gen LLMs, providing high-quality responses at low costs. And you can try it now!

Tanishq Mathew Abraham, Ph.D.

354,178 次观看 • 1 年前

LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models paper page: github: Recent advancements in text-to-image generation with diffusion models have yielded remarkable results synthesizing highly realistic and diverse images. However, these models still encounter difficulties when generating images from prompts that demand spatial or common sense reasoning. We propose to equip diffusion models with enhanced reasoning capabilities by using off-the-shelf pretrained large language models (LLMs) in a novel two-stage generation process. First, we adapt an LLM to be a text-guided layout generator through in-context learning. When provided with an image prompt, an LLM outputs a scene layout in the form of bounding boxes along with corresponding individual descriptions. Second, we steer a diffusion model with a novel controller to generate images conditioned on the layout. Both stages utilize frozen pretrained models without any LLM or diffusion model parameter optimization. We validate the superiority of our design by demonstrating its ability to outperform the base diffusion model in accurately generating images according to prompts that necessitate both language and spatial reasoning. Additionally, our method naturally allows dialog-based scene specification and is able to handle prompts in a language that is not well-supported by the underlying diffusion model.

LLM-grounded Diffusion: Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models paper page: github: Recent advancements in text-to-image generation with diffusion models have yielded remarkable results synthesizing highly realistic and diverse images. However, these models still encounter difficulties when generating images from prompts that demand spatial or common sense reasoning. We propose to equip diffusion models with enhanced reasoning capabilities by using off-the-shelf pretrained large language models (LLMs) in a novel two-stage generation process. First, we adapt an LLM to be a text-guided layout generator through in-context learning. When provided with an image prompt, an LLM outputs a scene layout in the form of bounding boxes along with corresponding individual descriptions. Second, we steer a diffusion model with a novel controller to generate images conditioned on the layout. Both stages utilize frozen pretrained models without any LLM or diffusion model parameter optimization. We validate the superiority of our design by demonstrating its ability to outperform the base diffusion model in accurately generating images according to prompts that necessitate both language and spatial reasoning. Additionally, our method naturally allows dialog-based scene specification and is able to handle prompts in a language that is not well-supported by the underlying diffusion model.

AK

83,657 次观看 • 2 年前

New episode of the Information Bottleneck! We talked with Stefano Ermon about why he thinks diffusion LLMs will replace autoregressive ones. Stefano co-invented DDIM, FlashAttention, DPO, and score-based diffusion models. He's a Stanford professor and now runs Inception AI, where they built Mercury II. We go deep but also cover the bigger picture - the startup journey, PhD vs industry, and where AI is heading. A few things that stuck with me: - He thinks of autoregressive models as typewriters and diffusion models as editors. One goes left to right. The other starts messy and refines. - Mercury II (their text difussion model) already beats the fastest autoregressive models on latency-critical stuff as voice agents, code suggestions, anything where you have a tight time budget. And it does it because diffusion generates tokens in parallel instead of one at a time. - We also got into whether AI will actually replace software engineers (his answer: no), PhD vs industry advice, and what it was like going from an ICML best paper to raising money.

New episode of the Information Bottleneck! We talked with Stefano Ermon about why he thinks diffusion LLMs will replace autoregressive ones. Stefano co-invented DDIM, FlashAttention, DPO, and score-based diffusion models. He's a Stanford professor and now runs Inception AI, where they built Mercury II. We go deep but also cover the bigger picture - the startup journey, PhD vs industry, and where AI is heading. A few things that stuck with me: - He thinks of autoregressive models as typewriters and diffusion models as editors. One goes left to right. The other starts messy and refines. - Mercury II (their text difussion model) already beats the fastest autoregressive models on latency-critical stuff as voice agents, code suggestions, anything where you have a tight time budget. And it does it because diffusion generates tokens in parallel instead of one at a time. - We also got into whether AI will actually replace software engineers (his answer: no), PhD vs industry advice, and what it was like going from an ICML best paper to raising money.

Ravid Shwartz Ziv

22,334 次观看 • 2 个月前

Try the first diffusion language model on the playground. Inception Labs' Mercury Coder generates over 1000 tokens/sec.

Try the first diffusion language model on the playground. Inception Labs' Mercury Coder generates over 1000 tokens/sec.

AI SDK

17,964 次观看 • 1 年前

Inception Labs just killed the transformer. They released Mercury 2, the world's first "diffusion" reasoning model. It's fast, and it uses a completely new model architecture... just watch this 11 min video to find out more:

Inception Labs just killed the transformer. They released Mercury 2, the world's first "diffusion" reasoning model. It's fast, and it uses a completely new model architecture... just watch this 11 min video to find out more:

David Ondrej

45,692 次观看 • 3 个月前

MotionStreamer announced on Hugging Face Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Space

MotionStreamer announced on Hugging Face Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Space

AK

43,387 次观看 • 1 年前

Testing out how well this interactive animated short made in Rive works on the ipad. Looks like I need to make a more touch friendly version, but the new hover feature with the pencil works great!

Testing out how well this interactive animated short made in Rive works on the ipad. Looks like I need to make a more touch friendly version, but the new hover feature with the pencil works great!

Drawsgood

45,567 次观看 • 2 年前

Hallucinator I wrote a gradio app to play with Standard Intelligence's Hertz model. Hertz inference works like LLM but purely audio based--the inference happens voice to voice. Hallucinator uses Hertz to autocomplete any speech clip, often resulting in some hilarious hallucinations.

Hallucinator I wrote a gradio app to play with Standard Intelligence's Hertz model. Hertz inference works like LLM but purely audio based--the inference happens voice to voice. Hallucinator uses Hertz to autocomplete any speech clip, often resulting in some hilarious hallucinations.

cocktail peanut

14,745 次观看 • 1 年前

Is any of this real? HeyGen approached me to try out their new Avatar IV model, at first I was a bit sceptical. But ohh I was quickly turned around. This is absolutely mad This is how it works:

Linus ✦ Ekenstam

176,792 次观看 • 1 年前

How to match the complexity of the problem you want to solve with the proper model. You want an inference router. In the video, I show you how simple and powerful this is. After this, you'll never talk directly to a model ever again.

How to match the complexity of the problem you want to solve with the proper model. You want an inference router. In the video, I show you how simple and powerful this is. After this, you'll never talk directly to a model ever again.

Santiago

11,861 次观看 • 19 天前

This is Audi's new supercar called the Nuvolari. I know it's a V8 hybrid, but I think it looks like an ugly EV! Am I wrong?

This is Audi's new supercar called the Nuvolari. I know it's a V8 hybrid, but I think it looks like an ugly EV! Am I wrong?

Vince Langman

14,634 次观看 • 8 天前

Just found the dLLM library to create Diffusion Language Models It's still early but it's insanely fun to experiment with diffusion (training, inference, eval) dLLM has the potential of becoming the main library for diffusion LLMs

Just found the dLLM library to create Diffusion Language Models It's still early but it's insanely fun to experiment with diffusion (training, inference, eval) dLLM has the potential of becoming the main library for diffusion LLMs

Maxime Labonne

75,421 次观看 • 7 个月前

Rewrote tiny-diffusion to be 3x smaller! Went from 951 lines to just 364, all contained in one file. As simple as possible, but not simpler. I also added a tiny GPT implementation as a comparison (312 lines, inspired by Andrej Karpathy). The two implementations are ~80% identical. The model architecture, training loop, tokenization, etc, only differ in 19 lines of code. The main differences are contained within two functions (generate and get_batch). The reason to include the GPT implementation was to show how similar autoregressive LMs are to diffusion LMs on an architectural level. Only *1* line of code in the architecture needs to be modified to support masked language diffusion instead of next-token prediction (by disabling causal masking). Link to the repo is in the comments

Rewrote tiny-diffusion to be 3x smaller! Went from 951 lines to just 364, all contained in one file. As simple as possible, but not simpler. I also added a tiny GPT implementation as a comparison (312 lines, inspired by Andrej Karpathy). The two implementations are ~80% identical. The model architecture, training loop, tokenization, etc, only differ in 19 lines of code. The main differences are contained within two functions (generate and get_batch). The reason to include the GPT implementation was to show how similar autoregressive LMs are to diffusion LMs on an architectural level. Only 1 line of code in the architecture needs to be modified to support masked language diffusion instead of next-token prediction (by disabling causal masking). Link to the repo is in the comments

Nathan Barry

161,578 次观看 • 5 个月前

I really want to know how the POLICE recruitment process works in this part of the world.

I really want to know how the POLICE recruitment process works in this part of the world.

YOUR EX 💔🌽🌽🌽🌽

616,066 次观看 • 2 个月前