Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

End to End Speech models are on fire - LLAMA-OMNI 8B - Apache licensed! 🔥 > Speech Encoder - Whisper Large v3 > LLM backbone - Llama 3.1 8B Instruct > Speech Decoder - HuBERT (UnitY) > Simultaneously generate Speech + Text > Less than 250 ms latency >... show more

Vaibhav (VB) Srivastav

50,123 subscribers

47,921 просмотров • 1 год назад •via X (Twitter)

Наука и технологии

Anya Rossi• Live Now

Private livecam show

Комментарии: 10

Фото профиля Vaibhav (VB) Srivastav

Vaibhav (VB) Srivastav1 год назад

Model checkpoint:

Фото профиля Vaibhav (VB) Srivastav

Vaibhav (VB) Srivastav1 год назад

Github repo:

Фото профиля Qingkai Fang

Qingkai Fang1 год назад

Thanks for sharing our work!

Фото профиля Vaibhav (VB) Srivastav

Vaibhav (VB) Srivastav1 год назад

🔥

Фото профиля Tommy D. Rossi

Tommy D. Rossi1 год назад

I wouldn't call this end to end, let's keep that term for single multi modal models that do everything by themselves

Фото профиля ThisAndThat

ThisAndThat1 год назад

less than 250ms latency on what?

Фото профиля Vaibhav (VB) Srivastav

Vaibhav (VB) Srivastav1 год назад

Time to first audio chunk according to their GH.

Фото профиля Waifuology

Waifuology1 год назад

License looks good, but the voice quality isn't really there yet.

Фото профиля Hiro

Hiro1 год назад

Do you know what are supported languages?

Фото профиля Trying my best :-)

Trying my best :-)1 год назад

Can it detect emotion?

Похожие видео

Multimodal Ichigo Llama 3.1 - Real Time Voice AI 🔥 > WhisperSpeech X Llama 3.1 8B > Trained on 50K hours of speech (7 languages) > Continually trained on 45hrs 10x A1000s > MLS -> WhisperVQ tokens -> Llama 3.1 > Instruction tuned on 1.89M samples > 70% speech, 20% transcription, 10% text > Apache 2.0 licensed ⚡ Architecture: > WhisperSpeech/ VQ for Semantic Tokens > Llama 3.1 8B Instruct for Text backbone > Early fusion (Chameleon) I'm super bullish on Homebrew to Menlo and early fusion, audio and text, multimodal models! (P.S. Play with the demo on Hugging Face)

Multimodal Ichigo Llama 3.1 - Real Time Voice AI 🔥 > WhisperSpeech X Llama 3.1 8B > Trained on 50K hours of speech (7 languages) > Continually trained on 45hrs 10x A1000s > MLS -> WhisperVQ tokens -> Llama 3.1 > Instruction tuned on 1.89M samples > 70% speech, 20% transcription, 10% text > Apache 2.0 licensed ⚡ Architecture: > WhisperSpeech/ VQ for Semantic Tokens > Llama 3.1 8B Instruct for Text backbone > Early fusion (Chameleon) I'm super bullish on Homebrew to Menlo and early fusion, audio and text, multimodal models! (P.S. Play with the demo on Hugging Face)

Vaibhav (VB) Srivastav

82,126 просмотров • 1 год назад

Pretty Insane - SoTA Text to Speech model capable of English AND Hindi - 3B Llama backbone - Apache 2.0 licensed 🔥 > Sub 80 ms latency > Supports both English, Hindi including code-mix > Runs in a free google colab too 🤯 Best part: They're actively working on other languages like Tamil, Telugu, Bengali, etc > Available on Hugging Face hub, powered by Transformers 💥

Pretty Insane - SoTA Text to Speech model capable of English AND Hindi - 3B Llama backbone - Apache 2.0 licensed 🔥 > Sub 80 ms latency > Supports both English, Hindi including code-mix > Runs in a free google colab too 🤯 Best part: They're actively working on other languages like Tamil, Telugu, Bengali, etc > Available on Hugging Face hub, powered by Transformers 💥

Vaibhav (VB) Srivastav

33,749 просмотров • 1 год назад

Wow! New Speech to Speech model - Fish Agent v0.1 3B by Fish Audio 🔥 > Trained on 700K hours of multilingual audio > Continue-pretrained version of Qwen-2.5-3B-Instruct for 200B audio & text tokens > Zero-shot voice cloning > Text + audio input/ Audio output > Ultra-fast inference w/ 200ms TTFA > Models on the Hub & Finetuning code on its way! 🚀 What an amazing time to be alive 🤗

Wow! New Speech to Speech model - Fish Agent v0.1 3B by Fish Audio 🔥 > Trained on 700K hours of multilingual audio > Continue-pretrained version of Qwen-2.5-3B-Instruct for 200B audio & text tokens > Zero-shot voice cloning > Text + audio input/ Audio output > Ultra-fast inference w/ 200ms TTFA > Models on the Hub & Finetuning code on its way! 🚀 What an amazing time to be alive 🤗

Vaibhav (VB) Srivastav

66,979 просмотров • 1 год назад

Today we’re releasing Chroma 1.0 → the world first open-source, end-to-end, real-time speech-to-speech model → with personalized voice cloning Trained by FlashLabs. Deployed on FlashAI👉 An open research-grade alternative to the OpenAI Realtime model. Voice Test dubbing Elon Musk and Lex Fridman: 🔥What’s real (evals and benchmarks attached): ⚡ 2× faster than real-time ) 📘 SGLang Cookbook: 📄 Paper + benchmarks: 🤗 Models: 💻 Inference code: 🔁 RT if you believe real-time AI should be open 💬 Reply if you’re building on Conversational Voice AI products

Today we’re releasing Chroma 1.0 → the world first open-source, end-to-end, real-time speech-to-speech model → with personalized voice cloning Trained by FlashLabs. Deployed on FlashAI👉 An open research-grade alternative to the OpenAI Realtime model. Voice Test dubbing Elon Musk and Lex Fridman: 🔥What’s real (evals and benchmarks attached): ⚡ 2× faster than real-time ) 📘 SGLang Cookbook: 📄 Paper + benchmarks: 🤗 Models: 💻 Inference code: 🔁 RT if you believe real-time AI should be open 💬 Reply if you’re building on Conversational Voice AI products

FlashLabs

1,248,841 просмотров • 6 месяцев назад

DeepSeek R1 671B is ridiculously good, but running at 158 tokens per second is something out of this world. A model *this good* running *this fast* is 3 years ahead of this time. The secret: DeepSeek in this short video isn't running on GPUs. It's running on custom-made chips! This is running on SambaNova and their SN40L chips, which are designed and optimized for running AI workflows. SambaNova is sponsoring my work. We designed GPUs to run games and later realized we could use them for AI. This is very different from creating a chip specifically for AI from the ground up. Just think about this for a second: One single SNL40 chip can simultaneously hold 100+ models (trillions of parameters) in memory! An agentic workflow that uses multiple models simultaneously won't need to swap models from memory even once! By the way, the Llama family of models *screams* on the SambaNova Cloud. Here are some numbers from my latest tests: • Llama 3.3 70B: 367.32 t/s • Llama 3.2 1B: 2381.82 t/s • Llama 3.2 3B: 1335.71 t/s • Llama 3.1 8B: 929.58 t/s • Llama 3.1 70B: 369.16 t/s • Llama 3.1 405B: 103.37 t/s You can access all of these models for free!

DeepSeek R1 671B is ridiculously good, but running at 158 tokens per second is something out of this world. A model this good running this fast is 3 years ahead of this time. The secret: DeepSeek in this short video isn't running on GPUs. It's running on custom-made chips! This is running on SambaNova and their SN40L chips, which are designed and optimized for running AI workflows. SambaNova is sponsoring my work. We designed GPUs to run games and later realized we could use them for AI. This is very different from creating a chip specifically for AI from the ground up. Just think about this for a second: One single SNL40 chip can simultaneously hold 100+ models (trillions of parameters) in memory! An agentic workflow that uses multiple models simultaneously won't need to swap models from memory even once! By the way, the Llama family of models screams on the SambaNova Cloud. Here are some numbers from my latest tests: • Llama 3.3 70B: 367.32 t/s • Llama 3.2 1B: 2381.82 t/s • Llama 3.2 3B: 1335.71 t/s • Llama 3.1 8B: 929.58 t/s • Llama 3.1 70B: 369.16 t/s • Llama 3.1 405B: 103.37 t/s You can access all of these models for free!

Santiago

131,019 просмотров • 1 год назад

Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale blog: Large-scale generative models such as GPT and DALL-E have revolutionized natural language processing and computer vision research. These models not only generate high fidelity text or image outputs, but are also generalists which can solve tasks not explicitly taught. In contrast, speech generative models are still primitive in terms of scale and task generalization. In this paper, we present Voicebox, the most versatile text-guided generative model for speech at scale. Voicebox is a non-autoregressive flow-matching model trained to infill speech, given audio context and text, trained on over 50K hours of speech that are neither filtered nor enhanced. Similar to GPT, Voicebox can perform many different tasks through in-context learning, but is more flexible as it can also condition on future context. Voicebox can be used for mono or cross-lingual zero-shot text-to-speech synthesis, noise removal, content editing, style conversion, and diverse sample generation. In particular, Voicebox outperforms the state-of-the-art zero-shot TTS model VALL-E on both intelligibility (5.9% vs 1.9% word error rates) and audio similarity (0.580 vs 0.681) while being up to 20 times faster.

Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale blog: Large-scale generative models such as GPT and DALL-E have revolutionized natural language processing and computer vision research. These models not only generate high fidelity text or image outputs, but are also generalists which can solve tasks not explicitly taught. In contrast, speech generative models are still primitive in terms of scale and task generalization. In this paper, we present Voicebox, the most versatile text-guided generative model for speech at scale. Voicebox is a non-autoregressive flow-matching model trained to infill speech, given audio context and text, trained on over 50K hours of speech that are neither filtered nor enhanced. Similar to GPT, Voicebox can perform many different tasks through in-context learning, but is more flexible as it can also condition on future context. Voicebox can be used for mono or cross-lingual zero-shot text-to-speech synthesis, noise removal, content editing, style conversion, and diverse sample generation. In particular, Voicebox outperforms the state-of-the-art zero-shot TTS model VALL-E on both intelligibility (5.9% vs 1.9% word error rates) and audio similarity (0.580 vs 0.681) while being up to 20 times faster.

AK

429,355 просмотров • 3 лет назад

How to build the world's fastest voice AI bot: - Self-host speech-to-text, LLM inference, and text-to-speech all together in the same container/cluster. - Route audio over the internet using WebRTC and edge networking. - Configure timings for voice activity detection, phrase endpointing, and other parts of the pipeline to optimize for latency. (There are trade-offs to doing this!) Here's a LLama 3 voice bot that has voice-to-voice response times of ~500ms. We used Deepgram's STT and TTS for this bot, and everything is hosted on cerebriumai's serverless GPU infrastructure.

How to build the world's fastest voice AI bot: - Self-host speech-to-text, LLM inference, and text-to-speech all together in the same container/cluster. - Route audio over the internet using WebRTC and edge networking. - Configure timings for voice activity detection, phrase endpointing, and other parts of the pipeline to optimize for latency. (There are trade-offs to doing this!) Here's a LLama 3 voice bot that has voice-to-voice response times of ~500ms. We used Deepgram's STT and TTS for this bot, and everything is hosted on cerebriumai's serverless GPU infrastructure.

kwindla

282,617 просмотров • 2 лет назад

Listen, speak, handle interruptions, and call tools in one live conversation. 🎙️ Introducing NVIDIA-NemotronLabs-VoiceChat-11B, NVIDIA’s end-to-end full-duplex model for real-time voice agents. 🤖 🏆 It ranks #2 among open full-duplex models on both VoiceBench and Full-Duplex-Bench 1.0. ⚡ Natural turn-taking responds in ~448 ms, while barge-in lets users interrupt the model with ~480 ms latency. 🛠️ The first open full-duplex model with live tool calling. One unified system handles speech understanding, generation, transcription, and tool execution while keeping the conversation flowing. 🧠 Built with a Fast Conformer, Nemotron Nano v2, and a TTS decoder, trained on ~550K hours of real and synthetic speech. 📜 Research use only. OpenMDW 1.1.

Listen, speak, handle interruptions, and call tools in one live conversation. 🎙️ Introducing NVIDIA-NemotronLabs-VoiceChat-11B, NVIDIA’s end-to-end full-duplex model for real-time voice agents. 🤖 🏆 It ranks #2 among open full-duplex models on both VoiceBench and Full-Duplex-Bench 1.0. ⚡ Natural turn-taking responds in ~448 ms, while barge-in lets users interrupt the model with ~480 ms latency. 🛠️ The first open full-duplex model with live tool calling. One unified system handles speech understanding, generation, transcription, and tool execution while keeping the conversation flowing. 🧠 Built with a Fast Conformer, Nemotron Nano v2, and a TTS decoder, trained on ~550K hours of real and synthetic speech. 📜 Research use only. OpenMDW 1.1.

ModelScope

21,207 просмотров • 6 дней назад

The latest MLX has a CUDA back-end! To get started: pip install "mlx[cuda]" With the same codebase you can develop locally, run your model on Apple silicon, or in the cloud on Nvidia GPUs. MLX is designed around Apple silicon - which has a unified memory architecture. It uses the same design with CUDA. So there's no need to move arrays around from CPU memory to GPU memory. Note, this is early days - some operations are missing and performance is still being optimized. But it's already quite fast for Transformer training, text generation, and more! Here's a demo using mlx-lm to generate text with Llama 3 8B (bf16) on an A100:

The latest MLX has a CUDA back-end! To get started: pip install "mlx[cuda]" With the same codebase you can develop locally, run your model on Apple silicon, or in the cloud on Nvidia GPUs. MLX is designed around Apple silicon - which has a unified memory architecture. It uses the same design with CUDA. So there's no need to move arrays around from CPU memory to GPU memory. Note, this is early days - some operations are missing and performance is still being optimized. But it's already quite fast for Transformer training, text generation, and more! Here's a demo using mlx-lm to generate text with Llama 3 8B (bf16) on an A100:

Awni Hannun

42,761 просмотров • 1 год назад

Voice agents are awkward, and everyone notices: You ask a question. The agent thinks. You wait. And wait... Nobody wants this. I'd rather talk to a person. If your model's response time is over 300ms, you won't make it. Unfortunately, most text-to-speech models can't get anywhere close to that. I want you to take a look at the latest model released by Inworld AI: TTS-1.5. I built a simple voice agent using the model so you can see it in action and test it on your computer. You'll find the repository link below. The latency numbers of this model are wild: • Max model → under 250ms • Mini model → under 130ms That's 4x faster than prior generations and faster than human response times!

Voice agents are awkward, and everyone notices: You ask a question. The agent thinks. You wait. And wait... Nobody wants this. I'd rather talk to a person. If your model's response time is over 300ms, you won't make it. Unfortunately, most text-to-speech models can't get anywhere close to that. I want you to take a look at the latest model released by Inworld AI: TTS-1.5. I built a simple voice agent using the model so you can see it in action and test it on your computer. You'll find the repository link below. The latency numbers of this model are wild: • Max model → under 250ms • Mini model → under 130ms That's 4x faster than prior generations and faster than human response times!

Santiago

69,444 просмотров • 6 месяцев назад

🔥HOLY SMOKES! $TAO holders! 🚀 SUBNET 19 (VISION) ON BITTENSOR IS ABSOLUTELY CRUSHING IT! In my 5+ years covering crypto and AI, this is one of the most impressive implementations I've seen. The combination of scale, performance, and decentralization is absolutely next level! 🚀 @namoray_dev @Corcel_X 💨 INSANE Speed Performance: - Llama 3.1 8B: 196.18 tokens/s with +107.23% advantage - Llama 3.1 70B: 124.96 tokens/s with +154.96% advantage - Llama 3.2 3B: 166.69 tokens/s with +21.66% advantage 🔥 Top Tier Model Integration: - Meta-Llama-3-70B & 8B Instruct - FLUX.1-schnell for Text-to-Image - ProteusV0.4-Lightning (Text & Image) - Multiple model variations for redundancy 🔥 What Makes This INSANE: - Complete decentralization - No single point of failure - Multiple model choices for redundancy - Real-time performance tracking - Transparent incentive structure The incentive distribution curve shows a healthy network with: - Strong rewards for top performers - Fair distribution across all participants - Clear path for growth and improvement - Sustainable economic model What's truly MIND-BLOWING is how they've managed to: 1. Scale to millions of operations 2. Maintain high quality across multiple tasks 3. Create a fair, competitive marketplace 4. Build in redundancy and reliability 5. Achieve true decentralization This isn't just another subnet - this is the future of decentralized AI inference happening RIGHT NOW! 🔥 1. MASSIVE Scale & Adoption: - We're seeing 7M+ tokens being processed - 14K+ processing steps being executed - Multiple AI models running simultaneously - Incredible miner participation across the network 2. Revolutionary Task Distribution: - Llama 3.1 70B leading with 20% weighting - Avatar Generation at 15% - Perfectly balanced task distribution for optimal network performance - Multiple specialized tasks including Text-to-Image and Image-to-Image processing 3. Elite Performance Metrics: - Top miners hitting 0.00775 incentive rates - Consistent performance across the network - Impressive scaling from top to bottom performers - Strong incentive curve maintaining network quality 📈 Network Performance: - Consistent upward trend in tokens/s - Quality scores maintaining high levels (>0.9) - Steady improvement in miner performance - Rock-solid network reliability ⚡ Platform Highlights: - Permissionless, serverless architecture - Global network of Always-On GPUs - Instant API access - Full decentralization - Multi-model support with seamless switching What makes this TRULY SPECIAL is the consistent upward trajectory in both speed and quality, while maintaining a decentralized architecture. The performance advantages over industry standards (+154.96% for 70B!) are absolutely mind-blowing! 🚀 This isn't just another AI subnet - it's a glimpse into the future of decentralized AI inference! The combination of speed, reliability, and model variety makes this one of the most impressive implementations in the space! 🔥 📽 Watch Now on YouTube and TikTok: Source 🔗

Andy ττ

11,616 просмотров • 1 год назад

Chinese robotics company Astribot released their latest World-Action Model (WAM), Lumo-2. Technical breakdown: - based on a frozen 🥶 Qwen-3.5 4B VLM - trained in 3 progressive stages: 1. Action is aligned with latent world dynamics (an abstract representation of action). Real-world actions are anchored to physical constraints, while the latent space is guided to focus on motion-relevant changes. This bidirectional relationship makes the model physically grounded -> critical for a world model. 2. Action is aligned with vision and language. Reusing the vision backbone and action encoder from the frozen VLM, the authors add a custom vocabulary (for new actions), a semantic module, an action decoder, and an action projector. This aligns the (new) action representations with the (existing) vision-language semantic space. Most importantly: it builds a direct mapping from natural-language instructions to motor execution. 3. End-to-end training on language, video, and robot data. Only the new modules (everything outside the frozen backbone) are trained end-to-end across temporal reasoning, physical understanding, long-horizon, and dexterous manipulation. At the end of the day, Lumo-2 is not the best on benchmarks, but that's not the point. What's genuinely new: - a way to combine latent world modeling and action generation through progressive alignment - a physically-grounded latent dynamics space - it lifts performance on unseen objects using un-annotated human egocentric video + Vision Pro captures, no special transfer algorithm needed Why it matters: - the whole model is thin trainable adapters (semantic module, action decoder/projector) on a frozen 4B backbone (cheap) - that scale is suited for real-time embedded inference (~2.71× decode speedup, no accuracy loss) - its real moat is long-horizon execution, where the added temporal memory pays off far more than on any other task As a result, this robot can now make your latte (5x sped up video):

Chinese robotics company Astribot released their latest World-Action Model (WAM), Lumo-2. Technical breakdown: - based on a frozen 🥶 Qwen-3.5 4B VLM - trained in 3 progressive stages: 1. Action is aligned with latent world dynamics (an abstract representation of action). Real-world actions are anchored to physical constraints, while the latent space is guided to focus on motion-relevant changes. This bidirectional relationship makes the model physically grounded -> critical for a world model. 2. Action is aligned with vision and language. Reusing the vision backbone and action encoder from the frozen VLM, the authors add a custom vocabulary (for new actions), a semantic module, an action decoder, and an action projector. This aligns the (new) action representations with the (existing) vision-language semantic space. Most importantly: it builds a direct mapping from natural-language instructions to motor execution. 3. End-to-end training on language, video, and robot data. Only the new modules (everything outside the frozen backbone) are trained end-to-end across temporal reasoning, physical understanding, long-horizon, and dexterous manipulation. At the end of the day, Lumo-2 is not the best on benchmarks, but that's not the point. What's genuinely new: - a way to combine latent world modeling and action generation through progressive alignment - a physically-grounded latent dynamics space - it lifts performance on unseen objects using un-annotated human egocentric video + Vision Pro captures, no special transfer algorithm needed Why it matters: - the whole model is thin trainable adapters (semantic module, action decoder/projector) on a frozen 4B backbone (cheap) - that scale is suited for real-time embedded inference (~2.71× decode speedup, no accuracy loss) - its real moat is long-horizon execution, where the added temporal memory pays off far more than on any other task As a result, this robot can now make your latte (5x sped up video):

Léo

32,078 просмотров • 18 дней назад

I'm running Llama 4 Maverick at 620 t/s! I'm living in the future! Honestly, a large language model running this fast is something straight out of a sci-fi movie. Speeds like this will enable a whole new world of applications that aren't possible today. For reference, GPT-4o, which is probably the most popular OpenAI model, runs between 60 and 110 t/s. The secret here: I'm not running AI at Meta's Llama 4 Maverick on a GPU. I'm using the SambaNova Cloud (my sponsor) and their custom SN40L chips. They are optimized from the ground up for running AI workflows. Right now, SambaNova Cloud runs DeepSeek, Qwen, Whisper, and the entire family of Llama models on these chips. You can check the speed of each of these models using SambaNova Cloud's Playground (see the attached video). It's completely free, and that's how I'm measuring their speeds. For example, I also tried DeepSeek R1 (the latest version from May) and, oh boy! DeepSeek R1 is a huge 671B parameter model. It's probably the best open reasoning model in the world, and it runs at 140 tokens per second! !!! Inference time on an SN40L is night and day from what you'll get from a GPU. Here is why this is big: If you are running an agentic workflow that uses multiple models simultaneously on a GPU, it will need to swap models in and out of memory (because not every model fits). A single SNL40 chip can simultaneously hold over 100 models (trillions of parameters) in memory. If you are using open models, try the SambaCloud API to see what lightning speed looks like. Here is how: 1. Create a free account at: 2. Check the QuickStart guide: If you try the playground, check the speed you're getting with Llama 4 and DeepSeek, and post the results below. I've seen much higher numbers than I posted here, so I'm curious to see whether geography affects the speed.

I'm running Llama 4 Maverick at 620 t/s! I'm living in the future! Honestly, a large language model running this fast is something straight out of a sci-fi movie. Speeds like this will enable a whole new world of applications that aren't possible today. For reference, GPT-4o, which is probably the most popular OpenAI model, runs between 60 and 110 t/s. The secret here: I'm not running AI at Meta's Llama 4 Maverick on a GPU. I'm using the SambaNova Cloud (my sponsor) and their custom SN40L chips. They are optimized from the ground up for running AI workflows. Right now, SambaNova Cloud runs DeepSeek, Qwen, Whisper, and the entire family of Llama models on these chips. You can check the speed of each of these models using SambaNova Cloud's Playground (see the attached video). It's completely free, and that's how I'm measuring their speeds. For example, I also tried DeepSeek R1 (the latest version from May) and, oh boy! DeepSeek R1 is a huge 671B parameter model. It's probably the best open reasoning model in the world, and it runs at 140 tokens per second! !!! Inference time on an SN40L is night and day from what you'll get from a GPU. Here is why this is big: If you are running an agentic workflow that uses multiple models simultaneously on a GPU, it will need to swap models in and out of memory (because not every model fits). A single SNL40 chip can simultaneously hold over 100 models (trillions of parameters) in memory. If you are using open models, try the SambaCloud API to see what lightning speed looks like. Here is how: 1. Create a free account at: 2. Check the QuickStart guide: If you try the playground, check the speed you're getting with Llama 4 and DeepSeek, and post the results below. I've seen much higher numbers than I posted here, so I'm curious to see whether geography affects the speed.

Santiago

34,148 просмотров • 1 год назад

The kyutai fully end-to-end audio model demo of today is a huge deal that many people missed in the room Mostly irrelevant are the facts that: - they come a few week after OpenAI ChatGPT-4o - the demo was less polished than the 4o one (in terms of voice quality, voice timing…) Relevant: - the model training pipeline and model archi are simple and hugely scalable, with a tiny 8+ people team like Kyutai building it in 4 months. Synthetic data is a huge enabler here - laser focus on local devices: Moshi will soon be everywhere. Frontier model builders have low incentive to let you run smaller models locally (price per token…) but non-profits like Kyutai have very different incentives. The Moshi demo is already online while the OpenAI 4o one is still in limbo. - going under 300 ms of latency while keeping Llama 8B or above quality of answers is a key enabler in terms of interactivity, it’s game changing, This feeling when the model answer your question before you even finished asking is quite crazy or when you interrupt the model while it’s talking and it react… Predictive coding in a model, instantly updated model of what you’re about to say... Basically they nailed the fundamentals. It’s here. This interactive voice tech will be everywhere. It will soon be an obvious commodity.

The kyutai fully end-to-end audio model demo of today is a huge deal that many people missed in the room Mostly irrelevant are the facts that: - they come a few week after OpenAI ChatGPT-4o - the demo was less polished than the 4o one (in terms of voice quality, voice timing…) Relevant: - the model training pipeline and model archi are simple and hugely scalable, with a tiny 8+ people team like Kyutai building it in 4 months. Synthetic data is a huge enabler here - laser focus on local devices: Moshi will soon be everywhere. Frontier model builders have low incentive to let you run smaller models locally (price per token…) but non-profits like Kyutai have very different incentives. The Moshi demo is already online while the OpenAI 4o one is still in limbo. - going under 300 ms of latency while keeping Llama 8B or above quality of answers is a key enabler in terms of interactivity, it’s game changing, This feeling when the model answer your question before you even finished asking is quite crazy or when you interrupt the model while it’s talking and it react… Predictive coding in a model, instantly updated model of what you’re about to say... Basically they nailed the fundamentals. It’s here. This interactive voice tech will be everywhere. It will soon be an obvious commodity.

Thomas Wolf

339,573 просмотров • 2 лет назад

people burn $1,900 a month on cloud, this $2,999 NVIDIA box runs 5 AI models at once and costs $10 to power 128GB of unified memory, stack two together and you double it, enough to run models a $2,000 graphics card cannot even open in the demo it ran Qwen3 Vision, Nemotron, a voice model, text to speech and speech to text all at once, and only used half the box here is the math that makes people switch: > cloud GPUs were costing $1,900 every single month, billed forever > the box is $2,999 one time, then about $10 of electricity a month > it paid for itself in two months, $22,000 stayed in the business in year one > zero data ever leaves your desk, no rental, no rate limits most people are still renting what this owns outright the full breakdown of the box and the exact math is in the article below

people burn $1,900 a month on cloud, this $2,999 NVIDIA box runs 5 AI models at once and costs $10 to power 128GB of unified memory, stack two together and you double it, enough to run models a $2,000 graphics card cannot even open in the demo it ran Qwen3 Vision, Nemotron, a voice model, text to speech and speech to text all at once, and only used half the box here is the math that makes people switch: > cloud GPUs were costing $1,900 every single month, billed forever > the box is $2,999 one time, then about $10 of electricity a month > it paid for itself in two months, $22,000 stayed in the business in year one > zero data ever leaves your desk, no rental, no rate limits most people are still renting what this owns outright the full breakdown of the box and the exact math is in the article below

John Doe

48,135 просмотров • 2 месяцев назад

Text-to-speech is moving way too fast. Just a few days ago, I tweeted about PersonaPlex-7B, NVIDIA's new open source TTS ( And today, Qwen just open-sourced Qwen3-TTS 🤯 It’s a revolutionary text-to-speech model built for control. Not just about generating speech, but about shaping how it sounds directly from language. You can guide the pace, the tone, and the expressiveness straight from text, without touching audio graphs or hand-tuning parameters. That’s the real shift! What makes Qwen3-TTS stand out is how practical it already is: → voice cloning from just a few seconds of audio → voice creation without any reference sample → support for 10 languages out of the box → end-to-end latency down to ~97ms → works in both streaming and non-streaming setups The models come in two sizes (0.6B and 1.7B), so you can trade off quality and hardware cost depending on your setup. You can work with curated voices, designed voices, or cloned ones, and it integrates cleanly with vLLM for production use. It also ships as a simple Python package you can pip install. If you’re building real-time voice systems, this removes a lot of friction! 100% free and open source. I put the repo in the 🧵↓

Text-to-speech is moving way too fast. Just a few days ago, I tweeted about PersonaPlex-7B, NVIDIA's new open source TTS ( And today, Qwen just open-sourced Qwen3-TTS 🤯 It’s a revolutionary text-to-speech model built for control. Not just about generating speech, but about shaping how it sounds directly from language. You can guide the pace, the tone, and the expressiveness straight from text, without touching audio graphs or hand-tuning parameters. That’s the real shift! What makes Qwen3-TTS stand out is how practical it already is: → voice cloning from just a few seconds of audio → voice creation without any reference sample → support for 10 languages out of the box → end-to-end latency down to ~97ms → works in both streaming and non-streaming setups The models come in two sizes (0.6B and 1.7B), so you can trade off quality and hardware cost depending on your setup. You can work with curated voices, designed voices, or cloned ones, and it integrates cleanly with vLLM for production use. It also ships as a simple Python package you can pip install. If you’re building real-time voice systems, this removes a lot of friction! 100% free and open source. I put the repo in the 🧵↓

Charly Wargnier

59,144 просмотров • 6 месяцев назад

QVAC SDK 0.15.0 is live. This release adds multiple prompts batching, brings a native AMD GPU backend to the stack, moves more vision encoders onto mobile GPUs, and adds a second local coding-agent integration. Main highlights: - Prompt batching for the LLM addon. Batch multiple prompts into one job and process them concurrently, with each answer returned the moment its generation finishes. - Native AMD GPU backend. A first-class HIP/ROCm backend in @qvac/vla-ggml, auto-selected over Vulkan with clean fallback when ROCm is absent. - A second local coding agent. OpenClaw joins OpenCode for local, cloud-free agent workflows. AGENTS - OpenCode plugin update (@qvac/opencode-plugin). Aligned with the current SDK, CLI, and AI SDK provider packages. A fresh install runs OpenCode against managed local QVAC models out of the box, from the default qvac/qwen3.5-9b, with no manual qvac serve setup. - OpenClaw plugin (@qvac/openclaw-plugin). A second coding-agent integration alongside OpenCode. A fresh setup installs the plugin, creates a local qvac provider through onboarding, and runs a QVAC model through OpenClaw🦞's local service path. LANGUAGE MODELS - Prompt batching (LLM addon). Batch multiple prompts in one job and run them concurrently, each answer returns the moment its generation finishes, no waiting on the others. - Reasoning-context trimming on hybrid + recurrent models (@qvac/llm-llamacpp). remove_thinking_from_context now works beyond pure-attention models. Same JS API, no throw. VOICE AND SPEECH - Transcription (transcription-parakeet 0.9.0). More robust CPU fallback on GPU failure and a faster Vulkan backend on Pixel 9. - Text-to-speech features (tts-ggml 0.4.0). Adds LavaSR for noise removal and adjustable output frequency up to 48 kHz, plus Japanese via Chatterbox. - Text-to-speech fixes (tts-ggml 0.4.1). CPU fallback on GPU failure, a q8_0 KV crash fix on Metal with Chatterbox. VISION - Qwen3.5 vision encoder on GPU (Android). Image encoder moves onto the phone GPU, with a smarter tile-grid preprocessor and default image-token caps, for flagship Android: Vulkan on Mali (Pixel 9 Pro) and OpenCL on Adreno 830 (Galaxy S25). - Gemma-4 vision encoder on GPU (Android). Vision encoder runs on the phone GPU instead of CPU, same flagship Android targets. PLATFORM AND PERFORMANCE - AMD GPU backend (@qvac/vla-ggml). Native HIP/ROCm backend, auto-selected over Vulkan with clean fallback when ROCm is absent (Linux x64 only). Comes with ~23% faster than Vulkan, ~14% faster than PyTorch-ROCm, parity preserved. Unified code style. A cleaner, more consistent, easier-to-contribute codebase. Let's build. npm install @qvac/sdk

QVAC SDK 0.15.0 is live. This release adds multiple prompts batching, brings a native AMD GPU backend to the stack, moves more vision encoders onto mobile GPUs, and adds a second local coding-agent integration. Main highlights: - Prompt batching for the LLM addon. Batch multiple prompts into one job and process them concurrently, with each answer returned the moment its generation finishes. - Native AMD GPU backend. A first-class HIP/ROCm backend in @qvac/vla-ggml, auto-selected over Vulkan with clean fallback when ROCm is absent. - A second local coding agent. OpenClaw joins OpenCode for local, cloud-free agent workflows. AGENTS - OpenCode plugin update (@qvac/opencode-plugin). Aligned with the current SDK, CLI, and AI SDK provider packages. A fresh install runs OpenCode against managed local QVAC models out of the box, from the default qvac/qwen3.5-9b, with no manual qvac serve setup. - OpenClaw plugin (@qvac/openclaw-plugin). A second coding-agent integration alongside OpenCode. A fresh setup installs the plugin, creates a local qvac provider through onboarding, and runs a QVAC model through OpenClaw🦞's local service path. LANGUAGE MODELS - Prompt batching (LLM addon). Batch multiple prompts in one job and run them concurrently, each answer returns the moment its generation finishes, no waiting on the others. - Reasoning-context trimming on hybrid + recurrent models (@qvac/llm-llamacpp). remove_thinking_from_context now works beyond pure-attention models. Same JS API, no throw. VOICE AND SPEECH - Transcription (transcription-parakeet 0.9.0). More robust CPU fallback on GPU failure and a faster Vulkan backend on Pixel 9. - Text-to-speech features (tts-ggml 0.4.0). Adds LavaSR for noise removal and adjustable output frequency up to 48 kHz, plus Japanese via Chatterbox. - Text-to-speech fixes (tts-ggml 0.4.1). CPU fallback on GPU failure, a q8_0 KV crash fix on Metal with Chatterbox. VISION - Qwen3.5 vision encoder on GPU (Android). Image encoder moves onto the phone GPU, with a smarter tile-grid preprocessor and default image-token caps, for flagship Android: Vulkan on Mali (Pixel 9 Pro) and OpenCL on Adreno 830 (Galaxy S25). - Gemma-4 vision encoder on GPU (Android). Vision encoder runs on the phone GPU instead of CPU, same flagship Android targets. PLATFORM AND PERFORMANCE - AMD GPU backend (@qvac/vla-ggml). Native HIP/ROCm backend, auto-selected over Vulkan with clean fallback when ROCm is absent (Linux x64 only). Comes with ~23% faster than Vulkan, ~14% faster than PyTorch-ROCm, parity preserved. Unified code style. A cleaner, more consistent, easier-to-contribute codebase. Let's build. npm install @qvac/sdk

QVAC

29,255,860 просмотров • 28 дней назад

Google just proved that bigger isn't always better. Their 308M parameter model is outperforming models 2x its size. Google just released 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝗚𝗲𝗺𝗺𝗮, and it's proving that lightweight embedding models can punch way above their weight class. At just 308M parameters (578MB), it's the new state-of-the-art for models under 500M parameters across MTEB multilingual, English, and code benchmarks. But the really impressive part is that it ranks 8th overall on MTEB(Multilingual, v2) - that's 𝟭𝟳 𝗽𝗹𝗮𝗰𝗲𝘀 above the second-best sub-500M model, and it's delivering performance 𝗰𝗼𝗺𝗽𝗮𝗿𝗮𝗯𝗹𝗲 𝘁𝗼 𝗺𝗼𝗱𝗲𝗹𝘀 𝗻𝗲𝗮𝗿𝗹𝘆 𝗱𝗼𝘂𝗯𝗹𝗲 𝗶𝘁𝘀 𝘀𝗶𝘇𝗲. There are three key parts of their training recipe that sets it apart: 𝟭. 𝗘𝗻𝗰𝗼𝗱𝗲𝗿-𝗗𝗲𝗰𝗼𝗱𝗲𝗿 𝗜𝗻𝗶𝘁𝗶𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Instead of starting from a decoder-only Gemma 3 model, they first adapted it to encoder-decoder, then used just the encoder. By basing EmbeddingGemma off an LLM that already has world and language understanding, it gives it a stronger starting point. 𝟮. 𝗧𝗵𝗿𝗲𝗲-𝗟𝗼𝘀𝘀 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 They combine three different loss functions, instead of just having one: • Contrastive loss (NCE) with in-batch negatives and hardness weighting • Spread-out regularization to ensure embeddings utilize the full space (for quantization and ANN retrieval) • Embedding matching distillation from Gemini Embedding - not just learning from relevance scores, but directly aligning the embedding space with the teacher model 𝟯. 𝗠𝗼𝗱𝗲𝗹 𝗦𝗼𝘂𝗽𝗶𝗻𝗴 Rather than just averaging checkpoints from the same training run, they use optimization techniques to find multiple specialized training mixtures. Each mixture creates an "expert" model in different domains, and averaging all their parameters creates a final model that's actually better than individual models. Extras: • Matryoshka embeddings supporting 768, 512, 256, and 128 dimensions • Quantization-aware training - maintains quality even at int4 precision • 100+ languages from Gemma 3 pretraining • Exceptional performance on low-resource languages (check their XTREME-UP results) Is it the absolute best embedding model? No - Gemini Embedding still leads overall. But that's not really the point. EmbeddingGemma proves you can achieve state-of-the-art performance in a small package that's actually deployable on-device, in low-latency applications, and in resource-constrained environments. This makes good embeddings accessible for use cases that I'm seeing more and more: offline applications, privacy-sensitive deployments, and high-throughput scenarios where inference cost actually matters. Full paper: Shoutout to the EmbeddingGemma team at Google DeepMind for this awesome open source work 💙 and to Daniel Williams for helping me with this video! 🫶

Google just proved that bigger isn't always better. Their 308M parameter model is outperforming models 2x its size. Google just released 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝗚𝗲𝗺𝗺𝗮, and it's proving that lightweight embedding models can punch way above their weight class. At just 308M parameters (578MB), it's the new state-of-the-art for models under 500M parameters across MTEB multilingual, English, and code benchmarks. But the really impressive part is that it ranks 8th overall on MTEB(Multilingual, v2) - that's 𝟭𝟳 𝗽𝗹𝗮𝗰𝗲𝘀 above the second-best sub-500M model, and it's delivering performance 𝗰𝗼𝗺𝗽𝗮𝗿𝗮𝗯𝗹𝗲 𝘁𝗼 𝗺𝗼𝗱𝗲𝗹𝘀 𝗻𝗲𝗮𝗿𝗹𝘆 𝗱𝗼𝘂𝗯𝗹𝗲 𝗶𝘁𝘀 𝘀𝗶𝘇𝗲. There are three key parts of their training recipe that sets it apart: 𝟭. 𝗘𝗻𝗰𝗼𝗱𝗲𝗿-𝗗𝗲𝗰𝗼𝗱𝗲𝗿 𝗜𝗻𝗶𝘁𝗶𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Instead of starting from a decoder-only Gemma 3 model, they first adapted it to encoder-decoder, then used just the encoder. By basing EmbeddingGemma off an LLM that already has world and language understanding, it gives it a stronger starting point. 𝟮. 𝗧𝗵𝗿𝗲𝗲-𝗟𝗼𝘀𝘀 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 They combine three different loss functions, instead of just having one: • Contrastive loss (NCE) with in-batch negatives and hardness weighting • Spread-out regularization to ensure embeddings utilize the full space (for quantization and ANN retrieval) • Embedding matching distillation from Gemini Embedding - not just learning from relevance scores, but directly aligning the embedding space with the teacher model 𝟯. 𝗠𝗼𝗱𝗲𝗹 𝗦𝗼𝘂𝗽𝗶𝗻𝗴 Rather than just averaging checkpoints from the same training run, they use optimization techniques to find multiple specialized training mixtures. Each mixture creates an "expert" model in different domains, and averaging all their parameters creates a final model that's actually better than individual models. Extras: • Matryoshka embeddings supporting 768, 512, 256, and 128 dimensions • Quantization-aware training - maintains quality even at int4 precision • 100+ languages from Gemma 3 pretraining • Exceptional performance on low-resource languages (check their XTREME-UP results) Is it the absolute best embedding model? No - Gemini Embedding still leads overall. But that's not really the point. EmbeddingGemma proves you can achieve state-of-the-art performance in a small package that's actually deployable on-device, in low-latency applications, and in resource-constrained environments. This makes good embeddings accessible for use cases that I'm seeing more and more: offline applications, privacy-sensitive deployments, and high-throughput scenarios where inference cost actually matters. Full paper: Shoutout to the EmbeddingGemma team at Google DeepMind for this awesome open source work 💙 and to Daniel Williams for helping me with this video! 🫶

Victoria Slocum

21,610 просмотров • 8 месяцев назад