Video yükleniyor...

Video Yüklenemedi

Bu video yüklenirken bir sorun oluştu. Bu geçici bir ağ sorunundan kaynaklanıyor olabilir veya video kullanılamıyor olabilir.

Ana Sayfaya Dön

Today's autoregressive models generate one token at a time. Mercury 2 generates tokens in parallel. Over 1,000 tok/sec on standard GPUs, at comparable quality to speed-optimized models. Since launch, the community has been showing what diffusion LLMs can unlock. Thanks to the team at Clyep for the breakdown.

Inception

18,261 subscribers

21,231 görüntüleme • 2 ay önce •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Yorum

Yorum bulunmuyor

Orijinal gönderinin yorumları burada görünecek

Benzer Videolar

What if language models didn't have to generate one token at a time? Our CEO Stefano Ermon joined @TBPN to break down how Mercury 2's diffusion LLM hits 1,000+ tok/s on standard NVIDIA GPUs — and why speed changes the product for coding, voice agents, and search.

What if language models didn't have to generate one token at a time? Our CEO Stefano Ermon joined @TBPN to break down how Mercury 2's diffusion LLM hits 1,000+ tok/s on standard NVIDIA GPUs — and why speed changes the product for coding, voice agents, and search.

Inception

27,108 görüntüleme • 4 ay önce

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

22,350 görüntüleme • 4 ay önce

“Next-token prediction” just got a serious rival 🤯 Ant Group just dropped LLaDA 2.1, and it challenges the dominant paradigm of LLMs. Unlike most models that generate one token at a time, LLaDA 2.1 uses diffusion to generate blocks of text in parallel. Why this changes everything: → Global Planning: It effectively sees the "future" while writing the "past" → Parallel Generation: It generates chunks in parallel, not sequentially, being much faster → Massive Efficiency: 16B MoE architecture that only uses ~1.4B active parameters per step. 100% Open Source.

“Next-token prediction” just got a serious rival 🤯 Ant Group just dropped LLaDA 2.1, and it challenges the dominant paradigm of LLMs. Unlike most models that generate one token at a time, LLaDA 2.1 uses diffusion to generate blocks of text in parallel. Why this changes everything: → Global Planning: It effectively sees the "future" while writing the "past" → Parallel Generation: It generates chunks in parallel, not sequentially, being much faster → Massive Efficiency: 16B MoE architecture that only uses ~1.4B active parameters per step. 100% Open Source.

Alvaro Cintas

31,277 görüntüleme • 5 ay önce

Excited to share what my team has been working on lately - Gemini diffusion! We bring diffusion to language modeling, yielding more power and blazing speeds! 🚀🚀🚀 Gemini diffusion is especially strong at coding. In this example the model generates at 2000 tokens/sec, including overheads like tokenization, prefill, safety filters etc.

Excited to share what my team has been working on lately - Gemini diffusion! We bring diffusion to language modeling, yielding more power and blazing speeds! 🚀🚀🚀 Gemini diffusion is especially strong at coding. In this example the model generates at 2000 tokens/sec, including overheads like tokenization, prefill, safety filters etc.

Brendan O'Donoghue

577,790 görüntüleme • 1 yıl önce

Our research team just released Flex-Forcing: a video generation method that lets a single model switch between generation methods at inference time. Right now there are two main approaches to video generation. Bidirectional diffusion models attend to every frame at once, holding structure well at the cost of speed. Autoregressive models generate frame by frame, so they stream fast and scale to long clips, but accumulate error and drift over time. Flex-Forcing trains a single model to do both, letting you choose from the range at inference based on your compute budget.

Our research team just released Flex-Forcing: a video generation method that lets a single model switch between generation methods at inference time. Right now there are two main approaches to video generation. Bidirectional diffusion models attend to every frame at once, holding structure well at the cost of speed. Autoregressive models generate frame by frame, so they stream fast and scale to long clips, but accumulate error and drift over time. Flex-Forcing trains a single model to do both, letting you choose from the range at inference based on your compute budget.

NVIDIA AI

31,412 görüntüleme • 22 gün önce

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, Ph.D. at ICML

354,254 görüntüleme • 1 yıl önce

Mercury 2 is live 🚀🚀 The world’s first reasoning diffusion LLM, delivering 5x faster performance than leading speed-optimized LLMs. Watching the team turn years of research into a real product never gets old, and I’m incredibly proud of what we’ve built. We’re just getting started on what diffusion can do for language.

Mercury 2 is live 🚀🚀 The world’s first reasoning diffusion LLM, delivering 5x faster performance than leading speed-optimized LLMs. Watching the team turn years of research into a real product never gets old, and I’m incredibly proud of what we’ve built. We’re just getting started on what diffusion can do for language.

Stefano Ermon

1,041,132 görüntüleme • 5 ay önce

Researchers found a way to make LLMs 8.5x faster! (without compromising accuracy) Speculative decoding is quite an effective way to address the single-token bottleneck in traditional LLM inference. A small "draft" model first generates the next several tokens, then the large model verifies all of them at once in a single forward pass. If a token at any position is wrong, you keep everything before it and restart from there. This never does worse than normal decoding. But current drafters in Speculative decoding still guess one token at a time. That makes the drafting step itself a bottleneck, capping real-world speedups at 2-3x. DFlash is a new technique that swaps the autoregressive drafter with a lightweight block diffusion model that guesses all tokens in one parallel shot. Drafting cost stays flat no matter how many tokens you speculate. On top of that, the drafter is conditioned on hidden features pulled from multiple layers of the target model and injected into every draft layer, so it makes significantly better guesses than a drafter working from scratch. In the side-by-side demo below, vanilla decoding runs at 48.5 tokens/sec. DFlash hits 415 tokens/sec on the same model, with zero quality loss. It's already integrated with vLLM, SGLang, and Transformers, with draft models on HuggingFace for several models like Qwen3, Qwen3.5, Llama 3.1, Kimi-K2.5, gpt-oss, and many more. I have shared the GitHub repo in the replies! KV caching is another must-know technique to boost LLM inference. I recently wrote an article about it. Read it below. 👉 Over to you: What use case are you working on that can benefit from this new technique?

Researchers found a way to make LLMs 8.5x faster! (without compromising accuracy) Speculative decoding is quite an effective way to address the single-token bottleneck in traditional LLM inference. A small "draft" model first generates the next several tokens, then the large model verifies all of them at once in a single forward pass. If a token at any position is wrong, you keep everything before it and restart from there. This never does worse than normal decoding. But current drafters in Speculative decoding still guess one token at a time. That makes the drafting step itself a bottleneck, capping real-world speedups at 2-3x. DFlash is a new technique that swaps the autoregressive drafter with a lightweight block diffusion model that guesses all tokens in one parallel shot. Drafting cost stays flat no matter how many tokens you speculate. On top of that, the drafter is conditioned on hidden features pulled from multiple layers of the target model and injected into every draft layer, so it makes significantly better guesses than a drafter working from scratch. In the side-by-side demo below, vanilla decoding runs at 48.5 tokens/sec. DFlash hits 415 tokens/sec on the same model, with zero quality loss. It's already integrated with vLLM, SGLang, and Transformers, with draft models on HuggingFace for several models like Qwen3, Qwen3.5, Llama 3.1, Kimi-K2.5, gpt-oss, and many more. I have shared the GitHub repo in the replies! KV caching is another must-know technique to boost LLM inference. I recently wrote an article about it. Read it below. 👉 Over to you: What use case are you working on that can benefit from this new technique?

Avi Chawla

157,390 görüntüleme • 2 ay önce

DFlash speculative decoding on Apple Silicon Qwen3.5-9B bf16 · M5 Max · greedy exact match ▸ 85 tok/s, 3.3× at 1024 tokens (runtime) ▸ ~70 tok/s, 2.6× in the video (terminal I/O overhead) ▸ 80 tok/s, 3.1× at 2048 tokens (runtime) Currently working on: → Long context (speedup degrades past 4K tokens, KV cache growth) → Int4 quantized models (27B class) Built on MLX, no CUDA, single machine. Draft generates 16 tokens in parallel, target verifies in one forward pass. Will open source when ready.

DFlash speculative decoding on Apple Silicon Qwen3.5-9B bf16 · M5 Max · greedy exact match ▸ 85 tok/s, 3.3× at 1024 tokens (runtime) ▸ ~70 tok/s, 2.6× in the video (terminal I/O overhead) ▸ 80 tok/s, 3.1× at 2048 tokens (runtime) Currently working on: → Long context (speedup degrades past 4K tokens, KV cache growth) → Int4 quantized models (27B class) Built on MLX, no CUDA, single machine. Draft generates 16 tokens in parallel, target verifies in one forward pass. Will open source when ready.

bstn 👁️

36,942 görüntüleme • 3 ay önce

Researchers at Meta recently shared MAGNeT, a single non-autoregressive transformer model for text-to-music & text-to-sound generation capable of generating audio on-par with the quality of SOTA models — at 7x the speed. MAGNeT is open source as part of AudioCraft. Hear audio samples and get more details on this work ➡️

Researchers at Meta recently shared MAGNeT, a single non-autoregressive transformer model for text-to-music & text-to-sound generation capable of generating audio on-par with the quality of SOTA models — at 7x the speed. MAGNeT is open source as part of AudioCraft. Hear audio samples and get more details on this work ➡️

AI at Meta

122,001 görüntüleme • 2 yıl önce

Scaling up GANs for Text-to-Image Synthesis present our 1B-parameter GigaGAN, achieving lower FID than Stable Diffusion v1.5, DALL·E 2, and Parti-750M. It generates 512px outputs at 0.13s, orders of magnitude faster than diffusion and autoregressive models, and inherits the disentangled, continuous, and controllable latent space of GANs abs: project page:

Scaling up GANs for Text-to-Image Synthesis present our 1B-parameter GigaGAN, achieving lower FID than Stable Diffusion v1.5, DALL·E 2, and Parti-750M. It generates 512px outputs at 0.13s, orders of magnitude faster than diffusion and autoregressive models, and inherits the disentangled, continuous, and controllable latent space of GANs abs: project page:

AK

278,115 görüntüleme • 3 yıl önce

Google isn’t betting on a single AI architecture. Sundar Pichai, CEO of Google: “We’re going to push the diffusion paradigm as hard as possible.” “All of today’s mainline Gemini models are autoregressive. Diffusion is a different paradigm.” “For the same capability, diffusion can be much faster.” “It’s behind the mainline models today, but there will be areas where it’s the right tool.” “We’re pushing multiple directions in parallel, and bringing them together where it makes sense.”

Google isn’t betting on a single AI architecture. Sundar Pichai, CEO of Google: “We’re going to push the diffusion paradigm as hard as possible.” “All of today’s mainline Gemini models are autoregressive. Diffusion is a different paradigm.” “For the same capability, diffusion can be much faster.” “It’s behind the mainline models today, but there will be areas where it’s the right tool.” “We’re pushing multiple directions in parallel, and bringing them together where it makes sense.”

Forward Future

152,919 görüntüleme • 7 ay önce

Hiring RL Engineer! Started off as a curious project at Lossfunk to push the boundaries of LLMs in social reasoning - we are now building RL environments, data, and benchmarks to simulate more real-world scenarios. If you want to train SoTA RL models over multi-GPUs (H200s/B200s) to unlock next AI frontier, this is for you.

Hiring RL Engineer! Started off as a curious project at Lossfunk to push the boundaries of LLMs in social reasoning - we are now building RL environments, data, and benchmarks to simulate more real-world scenarios. If you want to train SoTA RL models over multi-GPUs (H200s/B200s) to unlock next AI frontier, this is for you.

Satpal Singh Rathore

46,010 görüntüleme • 11 ay önce

🤯Topology & UV — the NO.1 headaches in #3D GenAI. 🔥We just move closer to BOTH at the same time. Introducing SATO: Strips as Tokens, a new autoregressive model for topology & UV, has been conditionally accepted to #SIGGRAPH 2026. Will available at #Hyper3D More details👇

🤯Topology & UV — the NO.1 headaches in #3D GenAI. 🔥We just move closer to BOTH at the same time. Introducing SATO: Strips as Tokens, a new autoregressive model for topology & UV, has been conditionally accepted to #SIGGRAPH 2026. Will available at #Hyper3D More details👇

Hyper3D by Deemos

87,787 görüntüleme • 3 ay önce

Ollama 0.2 is here! Concurrency is now enabled by default. This unlocks 2 major features: Parallel requests Ollama can now serve multiple requests at the same time, using only a little bit of additional memory for each request. This enables use cases such as: - Handling multiple chat sessions at the same time - Hosting code completion LLMs for your team - Processing different parts of a document simultaneously - Running multiple agents at the same time Run multiple models Ollama now supports loading different models at the same time. This improves several use cases: - Retrieval Augmented Generation (RAG): both the embedding and text completion models can be loaded into memory simultaneously. - Agents: multiple versions of an agent can now run simultaneously - Running large and small models side-by-side Models are automatically loaded and unloaded based on requests and how much GPU memory is available.

Ollama 0.2 is here! Concurrency is now enabled by default. This unlocks 2 major features: Parallel requests Ollama can now serve multiple requests at the same time, using only a little bit of additional memory for each request. This enables use cases such as: - Handling multiple chat sessions at the same time - Hosting code completion LLMs for your team - Processing different parts of a document simultaneously - Running multiple agents at the same time Run multiple models Ollama now supports loading different models at the same time. This improves several use cases: - Retrieval Augmented Generation (RAG): both the embedding and text completion models can be loaded into memory simultaneously. - Agents: multiple versions of an agent can now run simultaneously - Running large and small models side-by-side Models are automatically loaded and unloaded based on requests and how much GPU memory is available.

ollama

219,417 görüntüleme • 2 yıl önce

Images, video, audio, actions — generative modeling has converged on one recipe: compress into continuous latents, generate in latent space, decode. Language is the lone exception: still generated token by token, as a long discrete stream. Latent Thought Flows: We compress 256 text tokens into 8 continuous latents, generate them with a one-step flow model, and read them out as text using an autoregressive decoder. Result: a better inference-compute vs. generation-quality Pareto frontier than a tuned autoregressive baseline — thanks to compression and one-step generation. Co-led w/ Zhengyang Geng 🧵

Images, video, audio, actions — generative modeling has converged on one recipe: compress into continuous latents, generate in latent space, decode. Language is the lone exception: still generated token by token, as a long discrete stream. Latent Thought Flows: We compress 256 text tokens into 8 continuous latents, generate them with a one-step flow model, and read them out as text using an autoregressive decoder. Result: a better inference-compute vs. generation-quality Pareto frontier than a tuned autoregressive baseline — thanks to compression and one-step generation. Co-led w/ Zhengyang Geng 🧵

Mihir Prabhudesai

77,745 görüntüleme • 15 gün önce

At Hedra, we’re on a mission to make high-quality media creation accessible for everyone. By integrating fal into Hedra Studio, we’ve brought models like Minimax, Veo 2, and Kling into one platform so you can generate both images and videos in one platform.

At Hedra, we’re on a mission to make high-quality media creation accessible for everyone. By integrating fal into Hedra Studio, we’ve brought models like Minimax, Veo 2, and Kling into one platform so you can generate both images and videos in one platform.

Hedra

2,773,905 görüntüleme • 1 yıl önce

What a wonderful time with the product community in Türkiye! Thanks to the @useinsider team for hosting an A+ event at Boğaziçi University in Istanbul — the views from the event location were stunning and so was the community vibe. Can’t wait to be back.

What a wonderful time with the product community in Türkiye! Thanks to the @useinsider team for hosting an A+ event at Boğaziçi University in Istanbul — the views from the event location were stunning and so was the community vibe. Can’t wait to be back.

Shreyas Doshi

13,679 görüntüleme • 2 yıl önce

🇬🇧 “There’s over £1,000” “I believe it’s linked to drugs - we can size the money” Man in England is arrested for no reason other than he has a £1,000 on him in cash on him at the time.

🇬🇧 “There’s over £1,000” “I believe it’s linked to drugs - we can size the money” Man in England is arrested for no reason other than he has a £1,000 on him in cash on him at the time.

Concerned Citizen

121,758 görüntüleme • 1 yıl önce

🥂 ONE YEAR OF FAIR LAUNCHES 🥂 It’s been one year since MEME.COOKING went live on NEAR. From day one, our vision was simple: We set the table for something new on NEAR. A place where jokes could turn into tokens, banter could become a meme token, and ideas could grow into utility-driven memes. Over the past year, MEME.COOKING has become the #1 Decentralized Token Launchpad on NEAR, making it easy for anyone to launch a token. What makes it different is not just the simplicity, but the fairness, because here, everyone can cook, and anyone can dine. At the core, our mission has remained constant: fair launches only. One year in, and this is just the beginning. Here’s to more memes, more launches, and more fair tables set on NEAR.

🥂 ONE YEAR OF FAIR LAUNCHES 🥂 It’s been one year since MEME.COOKING went live on NEAR. From day one, our vision was simple: We set the table for something new on NEAR. A place where jokes could turn into tokens, banter could become a meme token, and ideas could grow into utility-driven memes. Over the past year, MEME.COOKING has become the #1 Decentralized Token Launchpad on NEAR, making it easy for anyone to launch a token. What makes it different is not just the simplicity, but the fairness, because here, everyone can cook, and anyone can dine. At the core, our mission has remained constant: fair launches only. One year in, and this is just the beginning. Here’s to more memes, more launches, and more fair tables set on NEAR.

MEME.COOKING

52,798 görüntüleme • 10 ay önce