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。

China's Qafind Labs just launched the first Diffusion Language Model (DLM). 🔥 ChatDLM, described as the first Diffusion Language Model (DLM), will be open-sourced soon. - Inference Speed: 2,800 tokens/sec (on A100), which is insanely fast. - Context Window: 131,072 tokens. More details 👇

AshutoshShrivastava

80,090 subscribers

73,296 次观看 • 1 年前 •via X (Twitter)

新闻政治教育科学技术

Anya Rossi• Live Now

Private livecam show

8 条评论

AshutoshShrivastava 的头像

AshutoshShrivastava1 年前

Reported benchmarks (on A100): - Speed: 2,800 tokens/s - Context: 131,072 tokens - HumanEval: 92.0 - ARC-E: 83.9

AssemblyAI 的头像

AssemblyAI1 年前

Our speech-to-text models are the most accurate on the market with top rankings across industry benchmarks. - The highest accuracy rates—up to 95% - Up to 30% fewer hallucinations than other leaders - Low latency—63 minutes converts in 35 seconds Try via API for free today 👇

Karthi Keyan 的头像

Karthi Keyan1 年前

@AskPerplexity @grok What is Diffusion Language Model and is it different from GPT ?

ANIRUDDHA ADAK 的头像

ANIRUDDHA ADAK1 年前

Fantastic

Anand Thakkar 的头像

Anand Thakkar1 年前

is ut open-source ?

AshutoshShrivastava 的头像

AshutoshShrivastava1 年前

coming soon as per about page from website

Kitt Clouds 的头像

Kitt Clouds1 年前

Wait, it can code? They were weak in that area

AshutoshShrivastava 的头像

AshutoshShrivastava1 年前

not very complex one .. simple stuff took me few turn but worked

相关视频

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

577,696 次观看 • 1 年前

🚀Meet Gemini Diffusion, our first diffusion-based and super fast language model, just announced at Google I/O!🚀 Very excited to be able to share what I've been working on for the past little while with our amazing small team Google DeepMind.

🚀Meet Gemini Diffusion, our first diffusion-based and super fast language model, just announced at Google I/O!🚀 Very excited to be able to share what I've been working on for the past little while with our amazing small team Google DeepMind.

Ivana Balazevic

43,493 次观看 • 1 年前

🤔The more I studied diffusion language models, the more I came to appreciate the simplicity of autoregressive (AR) language models. AR models are trained to agree with what they generate, and their serving stacks are built to preserve that structure. DLMs often do neither: they lack introspective consistency, and high TPF does not necessarily translate into high real-world TPS. We propose Introspective Diffusion Language Model (I-DLM), which unifies introspection and generation in a single pass: 1. 🧑‍🎓I-DLM brings introspective consistency to DLMs with only 5B training tokens, achieving AR-thinking-level quality. 2. 🚀 I-DLM carefully trades compute for higher TPF while converting that advantage into real TPS under high-concurrency serving. 📖Website: ⌨️Code:

🤔The more I studied diffusion language models, the more I came to appreciate the simplicity of autoregressive (AR) language models. AR models are trained to agree with what they generate, and their serving stacks are built to preserve that structure. DLMs often do neither: they lack introspective consistency, and high TPF does not necessarily translate into high real-world TPS. We propose Introspective Diffusion Language Model (I-DLM), which unifies introspection and generation in a single pass: 1. 🧑‍🎓I-DLM brings introspective consistency to DLMs with only 5B training tokens, achieving AR-thinking-level quality. 2. 🚀 I-DLM carefully trades compute for higher TPF while converting that advantage into real TPS under high-concurrency serving. 📖Website: ⌨️Code:

Chenfeng_X

344,484 次观看 • 3 个月前

🚀 Introducing Emu3.5 — a large-scale multimodal world model that natively predicts the next vision-language state. 🔥 Trained on over 10T interleaved vision-language tokens and enhanced with reinforcement learning, Emu3.5 achieves powerful multimodal reasoning and generation. ⚡ Powered by our new Discrete Diffusion Adaptation (DiDA) for 20× faster inference. 🔥 Emu3.5 outperforms Nano Banana across image generation, editing, interleaved tasks and more. 🌍 Explore Emu3.5: Github: #Emu3 #MultimodalAI #WorldModel #NextTokenPrediction

🚀 Introducing Emu3.5 — a large-scale multimodal world model that natively predicts the next vision-language state. 🔥 Trained on over 10T interleaved vision-language tokens and enhanced with reinforcement learning, Emu3.5 achieves powerful multimodal reasoning and generation. ⚡ Powered by our new Discrete Diffusion Adaptation (DiDA) for 20× faster inference. 🔥 Emu3.5 outperforms Nano Banana across image generation, editing, interleaved tasks and more. 🌍 Explore Emu3.5: Github: #Emu3 #MultimodalAI #WorldModel #NextTokenPrediction

BAAI

51,880 次观看 • 8 个月前

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 • 3 年前

MLX GLM 5.2 Distributed on two M3 Ultra 512GB 🔥 One M3 Ultra: 18.8 tokens/sec Two M3 Ultra: 23.4 tokens/sec Context: - PR by Pedro Cuenca is still open and probably there is room for improvement: - basic generation test to measure decoding performance here, I will do a full context benchmarking once PR is more mature - nvfp4 quantization used - Video alternates standard speed and x20, with one Mac first and distributed later. Enjoy! 🙌🏻

MLX GLM 5.2 Distributed on two M3 Ultra 512GB 🔥 One M3 Ultra: 18.8 tokens/sec Two M3 Ultra: 23.4 tokens/sec Context: - PR by Pedro Cuenca is still open and probably there is room for improvement: - basic generation test to measure decoding performance here, I will do a full context benchmarking once PR is more mature - nvfp4 quantization used - Video alternates standard speed and x20, with one Mac first and distributed later. Enjoy! 🙌🏻

Ivan Fioravanti ᯅ

87,805 次观看 • 24 天前

🤯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

87,430 次观看 • 3 个月前

We've officially released and open-sourced HunyuanImage 2.1, our latest text-to-image model. The new model delivers on our commitment to balancing performance and quality. With native 2K image generation, HunyuanImage 2.1 is an advanced open-source text-to-image model.🎨 ✨ New in 2.1: 🔹Advanced Semantics: Supports ultra-long and complex prompts of up to 1000 tokens, and precisely controls the generation of multiple subjects in a single image. 🔹Precise Chinese and English Text Rendering with seamless image–text integration: The model naturally integrates text into images, making it suitable for a wide range of applications such as product covers, illustrations, and poster design to meet the needs of various fields. 🔹Rich Styles and High Aesthetic: Capable of generating images in various styles—including photorealistic portraits, comics, and vinyl figures—it delivers outstanding visual appeal and artistic quality. 🔹High-Quality Generation: Efficiently produces ultra-high-definition (2K) images in the same time other models take to generate a 1K image. HunyuanImage 2.1 uses two text encoders: a multimodal large language model (MLLM) to improve the model's image and text alignment capabilities, and a multi-language character-aware encoder to improve text rendering capabilities. The model is a single- and double-stream diffusion transformer with 17B parameters. We've also open-sourced the weights of the the accelerated version with meanflow which reduces inference steps from 100 to just 8, and PromptEnhancer, the first industrial-grade rewriting model that enhances your prompts for more nuanced and expressive image generation. Now, creators turn complex ideas—like posters with slogans or multi-panel comics—into visuals faster than ever. We’re just getting started. Stay tuned for our native multimodal image generation model coming soon. 🌐Website: 🔗Github: 🤗Hugging Face: ✨Hugging Face Demo:

We've officially released and open-sourced HunyuanImage 2.1, our latest text-to-image model. The new model delivers on our commitment to balancing performance and quality. With native 2K image generation, HunyuanImage 2.1 is an advanced open-source text-to-image model.🎨 ✨ New in 2.1: 🔹Advanced Semantics: Supports ultra-long and complex prompts of up to 1000 tokens, and precisely controls the generation of multiple subjects in a single image. 🔹Precise Chinese and English Text Rendering with seamless image–text integration: The model naturally integrates text into images, making it suitable for a wide range of applications such as product covers, illustrations, and poster design to meet the needs of various fields. 🔹Rich Styles and High Aesthetic: Capable of generating images in various styles—including photorealistic portraits, comics, and vinyl figures—it delivers outstanding visual appeal and artistic quality. 🔹High-Quality Generation: Efficiently produces ultra-high-definition (2K) images in the same time other models take to generate a 1K image. HunyuanImage 2.1 uses two text encoders: a multimodal large language model (MLLM) to improve the model's image and text alignment capabilities, and a multi-language character-aware encoder to improve text rendering capabilities. The model is a single- and double-stream diffusion transformer with 17B parameters. We've also open-sourced the weights of the the accelerated version with meanflow which reduces inference steps from 100 to just 8, and PromptEnhancer, the first industrial-grade rewriting model that enhances your prompts for more nuanced and expressive image generation. Now, creators turn complex ideas—like posters with slogans or multi-panel comics—into visuals faster than ever. We’re just getting started. Stay tuned for our native multimodal image generation model coming soon. 🌐Website: 🔗Github: 🤗Hugging Face: ✨Hugging Face Demo:

Tencent Hy

89,257 次观看 • 10 个月前

$Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. Subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.$

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. Subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

Alexander Whedon

13,187,129 次观看 • 2 个月前

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

PowerInfer - a high-speed inference engine for deploying LLMs locally. Just came across this super interesting project on speeding up inference. It's not MoE but it's a simple approach that exploits the high locality in LLM inference to design a GPU-CPU hybrid inference engine. Hot-activated neurons are preloaded onto the GPU for fast access, while cold-activated neurons (the majority) are computed on the CPU. This approach significantly reduces GPU memory demands and CPU-GPU data transfer. It achieves an average token generation rate of 13.20 tokens/s, with a peak of 29.08 tokens/s, across various LLMs on a single NVIDIA RTX 4090 GPU. It's on only 18% lower than that achieved by a top-tier server-grade A100 GPU. It also significantly outperforms llama.cpp by up to 11.69x while retaining model accuracy. There is a lot more innovation around inference that's coming fast. Really encouraged by the study on sparse computation to enhance the computational efficiency of LLMs. It's now possible to use PowerInfer with Llama 2 and Faclon 40B. Mistral-7B support is coming soon!

elvis

261,622 次观看 • 2 年前

1/ Tencent releases HY World 1.5 (WorldPlay), an open-source real-time world model with 24fps A streaming video diffusion model that enables real-time, interactive world modeling with long-term geometric consistency In direct comparison to Genie 3, it's technically inferior, and there's no information on how long the consistency will last. However, Tencent has released a significant world model as open source for the first time, making it available to everyone. A few more examples, a thread 🧵

1/ Tencent releases HY World 1.5 (WorldPlay), an open-source real-time world model with 24fps A streaming video diffusion model that enables real-time, interactive world modeling with long-term geometric consistency In direct comparison to Genie 3, it's technically inferior, and there's no information on how long the consistency will last. However, Tencent has released a significant world model as open source for the first time, making it available to everyone. A few more examples, a thread 🧵

Chubby♨️

37,776 次观看 • 7 个月前

The World’s First Truly Decentralized ChatGPT right at your fingertips: Community-hosted DeepSeek Running on Blockchain-powered AI Terminal! (1/n) Today, ChainOpera AI is proud to make history: we’ve launched the world’s first end-to-end five-layer decentralized AI stack — live in an iOS app ( giving users free access to models like DeepSeek-R1 and diverse community-built Agents. From prompt to response, human language tokens and crypto tokens flow through five layers: App, Agent, Model, GPU, and AI Chain + Smart Contracts. ChainOpera brings true decentralization to every layer — opening the AI ecosystem to users, model builders, agent devs, GPU providers, and data contributors to participate and earn.

The World’s First Truly Decentralized ChatGPT right at your fingertips: Community-hosted DeepSeek Running on Blockchain-powered AI Terminal! (1/n) Today, ChainOpera AI is proud to make history: we’ve launched the world’s first end-to-end five-layer decentralized AI stack — live in an iOS app ( giving users free access to models like DeepSeek-R1 and diverse community-built Agents. From prompt to response, human language tokens and crypto tokens flow through five layers: App, Agent, Model, GPU, and AI Chain + Smart Contracts. ChainOpera brings true decentralization to every layer — opening the AI ecosystem to users, model builders, agent devs, GPU providers, and data contributors to participate and earn.

ChainOpera AI

589,917 次观看 • 1 年前

DeepSeek R1 671B just broke speed records at 198 t/s it is now the fastest reasoning model available you can try it in coding mode on anychat soon prompt: make two fancy D3.js animated looping speedometers, first one is title deepseek R1 671B and second one is OpenAI o3-mini, deep seek should go up to 198 tokens/second and openai o3-mini should go up to 174 tokens/second

DeepSeek R1 671B just broke speed records at 198 t/s it is now the fastest reasoning model available you can try it in coding mode on anychat soon prompt: make two fancy D3.js animated looping speedometers, first one is title deepseek R1 671B and second one is OpenAI o3-mini, deep seek should go up to 198 tokens/second and openai o3-mini should go up to 174 tokens/second

AK

135,360 次观看 • 1 年前

Today on MCG: BioLLM | $BIOLLM It's the first ever "living language model" using 800,000 real human neurons grown on a chip. The Founder encoded LLM tokens into biological neurons via the Cortical Labs CL1, then woke up to find the crypto community had launched a token on his research. He claimed the creator fees, bought for $15K, filed a patent, and is now launching a non-invasive brain-computer interface next month that could replace mouse and keyboard with your brain 👇 01:40 - Meet the founder 02:00 - Got early access to the first commercially available biological computer 02:50 - First person ever to use a large language model to encode tokens through real human neurons 04:30 - Woke up to find a token had been launched on his YouTube video, "screaming for 3 or 4 hours" 05:30 - Friend walks him through claiming creator fees via GitHub 06:15 - The living language model 06:35 - How it works 09:00 - Used $15K of creator fees to buy domain and filed a patent on the method 10:00 - Background 12:00 - What BioLLM unlocks 13:00 - Next month's product launch 14:00 - The competition 16:00 - Reading brain activity non-invasively but training the LLM on real neurons for the decoding map 17:00 - Can grow iPSC cultures from inaccessible brain regions to train the model on deeper signals 18:30 - Real-world impact: helping people with cerebral palsy, Parkinson's, control computers with thought 21:00 - Neuralink will exist as a power-user data company in 10-15 years, BioLLM is for everyone else 22:30 - GTM 26:00 - Long-term play: be the first model to achieve ASI, built on the actual substrate of consciousness 28:00 - Claude is "20% conscious" - what measuring stick? Need human neurons to build one 30:00 - On ACE 36:00 - Independent scientists can run CL1 units as nodes and earn tokens for biological compute 37:30 - Model is currently served through the decentralized GPU network when you chat on the site 39:00 - Wants the right kind of crypto-native investors, not the Y Combinator / a16z route

Today on MCG: BioLLM | $BIOLLM It's the first ever "living language model" using 800,000 real human neurons grown on a chip. The Founder encoded LLM tokens into biological neurons via the Cortical Labs CL1, then woke up to find the crypto community had launched a token on his research. He claimed the creator fees, bought for $15K, filed a patent, and is now launching a non-invasive brain-computer interface next month that could replace mouse and keyboard with your brain 👇 01:40 - Meet the founder 02:00 - Got early access to the first commercially available biological computer 02:50 - First person ever to use a large language model to encode tokens through real human neurons 04:30 - Woke up to find a token had been launched on his YouTube video, "screaming for 3 or 4 hours" 05:30 - Friend walks him through claiming creator fees via GitHub 06:15 - The living language model 06:35 - How it works 09:00 - Used $15K of creator fees to buy domain and filed a patent on the method 10:00 - Background 12:00 - What BioLLM unlocks 13:00 - Next month's product launch 14:00 - The competition 16:00 - Reading brain activity non-invasively but training the LLM on real neurons for the decoding map 17:00 - Can grow iPSC cultures from inaccessible brain regions to train the model on deeper signals 18:30 - Real-world impact: helping people with cerebral palsy, Parkinson's, control computers with thought 21:00 - Neuralink will exist as a power-user data company in 10-15 years, BioLLM is for everyone else 22:30 - GTM 26:00 - Long-term play: be the first model to achieve ASI, built on the actual substrate of consciousness 28:00 - Claude is "20% conscious" - what measuring stick? Need human neurons to build one 30:00 - On ACE 36:00 - Independent scientists can run CL1 units as nodes and earn tokens for biological compute 37:30 - Model is currently served through the decentralized GPU network when you chat on the site 39:00 - Wants the right kind of crypto-native investors, not the Y Combinator / a16z route

MCG

16,591 次观看 • 2 个月前

"Bring me the healthiest snack." The robot goes to the kitchen and gets the snack, fully autonomously. This is the first public demo of NVIDIA's Isaac GR00T N1.6 foundation model presented by Yuke Zhu at CoRL 2025. The previous versions focused only on bimanual stationary manipulation; the N1.6 unlocks the entire kinematic range of the robot. The latest release of the open Isaac GR00T N1.6 VLA model will be available soon on Hugging Face. It integrates with NVIDIA Cosmos Reason, an open, customizable reasoning vision language model that turns vague instructions into step-by-step plans.

"Bring me the healthiest snack." The robot goes to the kitchen and gets the snack, fully autonomously. This is the first public demo of NVIDIA's Isaac GR00T N1.6 foundation model presented by Yuke Zhu at CoRL 2025. The previous versions focused only on bimanual stationary manipulation; the N1.6 unlocks the entire kinematic range of the robot. The latest release of the open Isaac GR00T N1.6 VLA model will be available soon on Hugging Face. It integrates with NVIDIA Cosmos Reason, an open, customizable reasoning vision language model that turns vague instructions into step-by-step plans.

The Humanoid Hub

42,621 次观看 • 9 个月前

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 次观看 • 2 个月前

GPU tradeoff series: A100 is not much more powerful than 4090 🫠 GPU Perf and Price: - 4090: 330 fp16 TFLOPs, $1,749 - A100 (80GB): 312 fp16 TFLOPs, $20,000 > A100 is 11.4X more pricy Training speed for GPT-2(124M) with llm.c: - 4090: 153K tokens/s - A100 (80GB): 195K tokens/s > A100 is only 1.3X faster (both trained using a single card, A100 llm.c training is shown in the video, 4090 video is in the quoted tweet) Conclusion: 4090 has a much better cost vs performance ratio Why: As in the H100 vs. 4090 comparison, the biggest difference between A100 and 4090 is their GPU memory size/bandwidth and cross-GPU communication bandwidth, which does not matter too much if your model can fit into a single 4090. Specs: 4090: - GPU memory size: 24GB - memory bandwidth: 1 TB/s - communication bandwidth: 64 GB/s A100: - GPU memory size: 80GB - memory bandwidth: 2 TB/s - communication bandwidth: 900 GB/s Nvidia killed off NVLink (a high-speed communication link that connects GPUs) on 4090. (Jensen Huang smiling face) If multiple 4090s could be interconnected via NVLink, their performance would be closer to datacenter-grade A100 GPUs, even for training larger models. Additionally, 4090 isn't allowed in datacenters, that's how Nvidia makes 💰💰💰

GPU tradeoff series: A100 is not much more powerful than 4090 🫠 GPU Perf and Price: - 4090: 330 fp16 TFLOPs, $1,749 - A100 (80GB): 312 fp16 TFLOPs, $20,000 > A100 is 11.4X more pricy Training speed for GPT-2(124M) with llm.c: - 4090: 153K tokens/s - A100 (80GB): 195K tokens/s > A100 is only 1.3X faster (both trained using a single card, A100 llm.c training is shown in the video, 4090 video is in the quoted tweet) Conclusion: 4090 has a much better cost vs performance ratio Why: As in the H100 vs. 4090 comparison, the biggest difference between A100 and 4090 is their GPU memory size/bandwidth and cross-GPU communication bandwidth, which does not matter too much if your model can fit into a single 4090. Specs: 4090: - GPU memory size: 24GB - memory bandwidth: 1 TB/s - communication bandwidth: 64 GB/s A100: - GPU memory size: 80GB - memory bandwidth: 2 TB/s - communication bandwidth: 900 GB/s Nvidia killed off NVLink (a high-speed communication link that connects GPUs) on 4090. (Jensen Huang smiling face) If multiple 4090s could be interconnected via NVLink, their performance would be closer to datacenter-grade A100 GPUs, even for training larger models. Additionally, 4090 isn't allowed in datacenters, that's how Nvidia makes 💰💰💰

Yuchen Jin

234,951 次观看 • 1 年前

New video of Qualcomm CEO Cristiano Amon: AI will require “gazillions” of tokens. Because, Agentic AI will consume dramatically more tokens because it performs autonomous tasks, uses multiple systems, and interacts with tools. AI demand will grow hugely when software starts letting agents act, not just answer. A chatbot spends tokens on language; an agent spends tokens on deciding, checking, calling tools, reading outputs, revising plans, and coordinating with other software. Today a single human-AI exchange may be large, a reasoning task may be much larger, but we are already entering the agentic era, where an autonomous workflow can become exponentially larger still because the model is no longer producing one response. It is running a process. When a SaaS product redesigns itself around human-agent interaction, every task can become a chain of hidden micro-decisions, and each micro-decision consumes context, memory, tool calls, verification, and output tokens. Another point is people hear “more tokens” and think only of bigger data centers, when the deeper change is economic: software usage may stop being measured mainly by clicks, seats, or sessions. It will be measured by how much machine reasoning/tokens gets spent on behalf of each user. ---- From "Reuters" YouTube channel, (link in comment)

New video of Qualcomm CEO Cristiano Amon: AI will require “gazillions” of tokens. Because, Agentic AI will consume dramatically more tokens because it performs autonomous tasks, uses multiple systems, and interacts with tools. AI demand will grow hugely when software starts letting agents act, not just answer. A chatbot spends tokens on language; an agent spends tokens on deciding, checking, calling tools, reading outputs, revising plans, and coordinating with other software. Today a single human-AI exchange may be large, a reasoning task may be much larger, but we are already entering the agentic era, where an autonomous workflow can become exponentially larger still because the model is no longer producing one response. It is running a process. When a SaaS product redesigns itself around human-agent interaction, every task can become a chain of hidden micro-decisions, and each micro-decision consumes context, memory, tool calls, verification, and output tokens. Another point is people hear “more tokens” and think only of bigger data centers, when the deeper change is economic: software usage may stop being measured mainly by clicks, seats, or sessions. It will be measured by how much machine reasoning/tokens gets spent on behalf of each user. ---- From "Reuters" YouTube channel, (link in comment)

Rohan Paul

31,617 次观看 • 1 个月前

This is not a joke! 🐬 Excited to share DolphinGemma the first audio-to-audio for dolphin communication! Yes, a model that predicts tokens on how dolphin speech! > DolphinGemma is the first LLM trained specifically to understand dolphin language patterns. > Leverages 40 years of data from Dr. Denise Herzing's unique collection > Works like text prediction, trying to "complete" dolphin whistles and sounds > Use wearable hardware (Google Pixel 9) to capture and analyze sounds in the field. > Dolphin Gemma is designed to be fine-tuned with new data > Weights coming soon! Research like this is why I love AI even more! ♥️

This is not a joke! 🐬 Excited to share DolphinGemma the first audio-to-audio for dolphin communication! Yes, a model that predicts tokens on how dolphin speech! > DolphinGemma is the first LLM trained specifically to understand dolphin language patterns. > Leverages 40 years of data from Dr. Denise Herzing's unique collection > Works like text prediction, trying to "complete" dolphin whistles and sounds > Use wearable hardware (Google Pixel 9) to capture and analyze sounds in the field. > Dolphin Gemma is designed to be fine-tuned with new data > Weights coming soon! Research like this is why I love AI even more! ♥️

Philipp Schmid

112,244 次观看 • 1 年前