Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Red Hat AI just shipped DFlash speculator checkpoints for two of NVIDIA AI's most powerful open models: → Nemotron Ultra 550B → Nemotron Super 120B On math and reasoning: ~5 out of 7 draft tokens accepted on average. On code (HumanEval): ~3.4 out of 7. Both checkpoints trained with... show more

Red Hat AI

12,733 subscribers

15,077 views • 2 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

4 Comments

Helen Zhao2 months ago

@NVIDIAAI Training compute for this model was generously provided by @LambdaAPI, a leading cloud platform built for AI training and inference. Thanks again @TheZachMueller for all the help and support along the way!

RiftStack AI2 months ago

@NVIDIAAI any intuition why code accepts so many fewer draft tokens than math?

Oli Wilkins2 months ago

@NVIDIAAI How does this compare to something like EAGLE?

Mike Gannotti2 months ago

@NVIDIAAI I’m really looking forward to seeing what NVIDIA drops when Nemotron hits 4. 3 is already such a strong set of models

Related Videos

Jensen on Nemotron 3 Ultra and it's role in the open ecosystem. It's great to have Nvidia leading on this, with so much turnover in open model providers in recent years. Super excited to see how the model turns out.

Jensen on Nemotron 3 Ultra and it's role in the open ecosystem. It's great to have Nvidia leading on this, with so much turnover in open model providers in recent years. Super excited to see how the model turns out.

Nathan Lambert

26,650 views • 6 months ago

🤝 Arcee.ai, an American AI lab, built its frontier open model family on NVIDIA Blackwell Ultra. Its flagship MoE model, Trinity-Large-Thinking, was RL post-trained using NVIDIA NeMo open libraries. Arcee optimized agentic model inference with NVIDIA Dynamo and vLLM, along with NVIDIA accelerated networking, to deliver low token cost, achieving: ✅ 3T+ tokens served on OpenRouter in its first two months ✅ $0.90 per 1M output tokens ✅ 2nd on PinchBench for open agentic models Explore Arcee's open models, powered by NVIDIA ➡️

🤝 Arcee.ai, an American AI lab, built its frontier open model family on NVIDIA Blackwell Ultra. Its flagship MoE model, Trinity-Large-Thinking, was RL post-trained using NVIDIA NeMo open libraries. Arcee optimized agentic model inference with NVIDIA Dynamo and vLLM, along with NVIDIA accelerated networking, to deliver low token cost, achieving: ✅ 3T+ tokens served on OpenRouter in its first two months ✅ $0.90 per 1M output tokens ✅ 2nd on PinchBench for open agentic models Explore Arcee's open models, powered by NVIDIA ➡️

NVIDIA AI Infrastructure

27,377 views • 2 months ago

We are excited to join Nvidia's Nemotron Coalition of leading AI labs working together to advance open frontier foundation models. To celebrate we have partnered with NVIDIA and Nebius Token Factory to provide 2 free weeks of the new Nemotron 3 Ultra model on the Nous Portal!

We are excited to join Nvidia's Nemotron Coalition of leading AI labs working together to advance open frontier foundation models. To celebrate we have partnered with NVIDIA and Nebius Token Factory to provide 2 free weeks of the new Nemotron 3 Ultra model on the Nous Portal!

Nous Research

1,501,926 views • 4 months ago

NVIDIA is giving away free access to 140+ AI models for a full year > most people building AI agents are paying $50-200/month for API access NVIDIA just made that argument irrelevant models you get: GLM 5.2, MiniMax M3, Nemotron-3-Ultra-550B-A55B, Kimi K2.7, and 130+ more setup: > step 1 - get your free key > go to > register -> bind phone -> copy API key > step 2 - add to Hermes agent > open Settings -> Model Provider -> Custom base_url = " api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx" > step 3 - pick a model model = "z-ai/glm-5.2" model = "minimaxai/minimax-m3" model = "nvidia/nemotron-3-ultra-550b-a55b" model = "moonshot-ai/kimi-k2.7" > Hermes already has NVIDIA set as default base_url > paste the key and you're running instantly > works the same in Cursor and OpenCode > cost: $0 > limit: 40 req/min > expires: 1 year while everyone is paying for API access, this is sitting there for free with this many models in one place, it is also the easiest way to test which one actually fits your agent before you commit to a paid provider

NVIDIA is giving away free access to 140+ AI models for a full year > most people building AI agents are paying $50-200/month for API access NVIDIA just made that argument irrelevant models you get: GLM 5.2, MiniMax M3, Nemotron-3-Ultra-550B-A55B, Kimi K2.7, and 130+ more setup: > step 1 - get your free key > go to > register -> bind phone -> copy API key > step 2 - add to Hermes agent > open Settings -> Model Provider -> Custom base_url = " api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx" > step 3 - pick a model model = "z-ai/glm-5.2" model = "minimaxai/minimax-m3" model = "nvidia/nemotron-3-ultra-550b-a55b" model = "moonshot-ai/kimi-k2.7" > Hermes already has NVIDIA set as default base_url > paste the key and you're running instantly > works the same in Cursor and OpenCode > cost: $0 > limit: 40 req/min > expires: 1 year while everyone is paying for API access, this is sitting there for free with this many models in one place, it is also the easiest way to test which one actually fits your agent before you commit to a paid provider

Mr. Buzzoni

121,850 views • 2 months ago

> be Nemotron 3 Ultra > announced five days ago > still full alpha > 550B parameters > open weights -> free tier -> 1,000,000 tokens of context > paste 30 competitors in one block -> it reads all of it > 94.7 recall at a million tokens > 78.7 truthfulness -> best in the set > $0.50 in -> $2.50 out -> per million > everyone still rents closed GPT for this > NVIDIA shipped it for pennies -> five days ago > the smartest AI gets the applause > the cheapest one that works gets paid > and you're this early

> be Nemotron 3 Ultra > announced five days ago > still full alpha > 550B parameters > open weights -> free tier -> 1,000,000 tokens of context > paste 30 competitors in one block -> it reads all of it > 94.7 recall at a million tokens > 78.7 truthfulness -> best in the set > $0.50 in -> $2.50 out -> per million > everyone still rents closed GPT for this > NVIDIA shipped it for pennies -> five days ago > the smartest AI gets the applause > the cheapest one that works gets paid > and you're this early

shmidt

20,200 views • 4 months ago

NVIDIA is giving away free access to 130+ AI models for a full year > most people building AI agents are paying $50-200/month for API access NVIDIA just made that argument irrelevant models you get: MiniMax M2.7, GLM 5.1, Kimi 2.5, DeepSeek-v4-flash, GPT-OSS-120B and 110+ more setup: > step 1 - get your free key > go to > register -> bind phone -> copy API key > step 2 - add to Hermes agent > open Settings -> Model Provider -> Custom base_url = " api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx" > step 3 - pick a model model = "minimaxai/minimax-m2.7" model = "zhipuai/glm-5.1" model = "moonshot-ai/kimi-2.5" model = "deepseek-ai/deepseek-v4-flash" model = "nvidia/nemotron-3-ultra-550b-a55b" > Hermes already has NVIDIA set as default base_url > paste the key and you're running instantly > works the same in Cursor and OpenCode > cost: $0 > limit: 40 req/min > expires: 1 year while everyone is paying for API access, this is sitting there for free

NVIDIA is giving away free access to 130+ AI models for a full year > most people building AI agents are paying $50-200/month for API access NVIDIA just made that argument irrelevant models you get: MiniMax M2.7, GLM 5.1, Kimi 2.5, DeepSeek-v4-flash, GPT-OSS-120B and 110+ more setup: > step 1 - get your free key > go to > register -> bind phone -> copy API key > step 2 - add to Hermes agent > open Settings -> Model Provider -> Custom base_url = " api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx" > step 3 - pick a model model = "minimaxai/minimax-m2.7" model = "zhipuai/glm-5.1" model = "moonshot-ai/kimi-2.5" model = "deepseek-ai/deepseek-v4-flash" model = "nvidia/nemotron-3-ultra-550b-a55b" > Hermes already has NVIDIA set as default base_url > paste the key and you're running instantly > works the same in Cursor and OpenCode > cost: $0 > limit: 40 req/min > expires: 1 year while everyone is paying for API access, this is sitting there for free

Mr. Buzzoni

196,730 views • 4 months ago

Inside Nemotron and NVIDIA's AI lab: my conversation with Bryan Catanzaro (Bryan Catanzaro). NVIDIA is a chip company. So why does it put hundreds of researchers on building AI models - and then give them away for free? We go deep into the Nemotron models, what it takes to build a top AI lab, and the future of frontier AI. 01:33 - Is open source AI catching the frontier? 05:29 - Do closed labs blocking distillation slow open source down? 07:42 - Is the US falling behind China? 10:30 - Why companies actually choose open models 12:39 - A "crazy" 2008 bet: machine learning on GPUs 15:33 - Working with Andrew Ng and Dario Amodei at Baidu 17:41 - Coming back to NVIDIA: DLSS and the birth of Megatron 21:55 - The real reason NVIDIA builds its own models 24:28 - Is Moore's Law really dead? 33:37 - The Nemotron family: Nano, Super, Ultra 35:09 - Built for agents: why NVIDIA bets on speed 36:02 - How you train a 550B model in 4 bits 39:25 - Hybrid Mamba-Transformer, explained simply 42:31 - Mixture of experts, and why NVIDIA built NVL72 around it 47:26 - Why a 1-million-token context window matters 49:26 - Multi-token prediction: how the model predicts 5 tokens at once 52:47 - Multi-teacher distillation: teaching one model from many 58:01 - Where reinforcement learning goes next 01:00:16 - Inside NVIDIA's research org: "the mission is the boss" 01:04:03 - How NVIDIA decides who gets the GPUs 01:10:53 - Why NVIDIA still feels entrepreneurial after 33 years 01:12:58 - Why Bryan doesn't believe in the singularity 01:17:50 - The AI backlash 01:19:18 - The controversial case: open AI is safer than closed

Inside Nemotron and NVIDIA's AI lab: my conversation with Bryan Catanzaro (Bryan Catanzaro). NVIDIA is a chip company. So why does it put hundreds of researchers on building AI models - and then give them away for free? We go deep into the Nemotron models, what it takes to build a top AI lab, and the future of frontier AI. 01:33 - Is open source AI catching the frontier? 05:29 - Do closed labs blocking distillation slow open source down? 07:42 - Is the US falling behind China? 10:30 - Why companies actually choose open models 12:39 - A "crazy" 2008 bet: machine learning on GPUs 15:33 - Working with Andrew Ng and Dario Amodei at Baidu 17:41 - Coming back to NVIDIA: DLSS and the birth of Megatron 21:55 - The real reason NVIDIA builds its own models 24:28 - Is Moore's Law really dead? 33:37 - The Nemotron family: Nano, Super, Ultra 35:09 - Built for agents: why NVIDIA bets on speed 36:02 - How you train a 550B model in 4 bits 39:25 - Hybrid Mamba-Transformer, explained simply 42:31 - Mixture of experts, and why NVIDIA built NVL72 around it 47:26 - Why a 1-million-token context window matters 49:26 - Multi-token prediction: how the model predicts 5 tokens at once 52:47 - Multi-teacher distillation: teaching one model from many 58:01 - Where reinforcement learning goes next 01:00:16 - Inside NVIDIA's research org: "the mission is the boss" 01:04:03 - How NVIDIA decides who gets the GPUs 01:10:53 - Why NVIDIA still feels entrepreneurial after 33 years 01:12:58 - Why Bryan doesn't believe in the singularity 01:17:50 - The AI backlash 01:19:18 - The controversial case: open AI is safer than closed

Matt Turck

56,954 views • 3 months ago

NVIDIA just released a new open source transcription model, Nemotron Speech ASR, designed from the ground up for low-latency use cases like voice agents. Here's a voice agent built with this new model. 24ms transcription finalization and total voice-to-voice inference time under 500ms. This agent actually uses *three* NVIDIA open source models: - Nemotron Speech ASR - Nemotron 3 Nano 30GB in a 4-bit quant (released in December) - A preview checkpoint of the upcoming Magpie text-to-speech model These models are all truly open source: weights, training data, training code, and inference code. This is a big deal! Jensen said in the CES keynote yesterday that he expects open source models to catch up to proprietary models this year in a number of categories. NVIDIA is putting their weight behind making this happen. (As Alan Kay said, the best way to predict the future is to invent it.) The code for this agent is open source too, of course. You can deploy it to production with Modal and Pipecat AI cloud, or run locally on an NVIDIA DGX Spark or RTX 5090.

NVIDIA just released a new open source transcription model, Nemotron Speech ASR, designed from the ground up for low-latency use cases like voice agents. Here's a voice agent built with this new model. 24ms transcription finalization and total voice-to-voice inference time under 500ms. This agent actually uses three NVIDIA open source models: - Nemotron Speech ASR - Nemotron 3 Nano 30GB in a 4-bit quant (released in December) - A preview checkpoint of the upcoming Magpie text-to-speech model These models are all truly open source: weights, training data, training code, and inference code. This is a big deal! Jensen said in the CES keynote yesterday that he expects open source models to catch up to proprietary models this year in a number of categories. NVIDIA is putting their weight behind making this happen. (As Alan Kay said, the best way to predict the future is to invent it.) The code for this agent is open source too, of course. You can deploy it to production with Modal and Pipecat AI cloud, or run locally on an NVIDIA DGX Spark or RTX 5090.

kwindla

274,867 views • 9 months ago

Hot off the press 🔥 AEON vLLM Ultimate → v0.24.0. Link below ⤵️ The ultimate container for AI on NVIDIA DGX Spark — vLLM compiled from source for GB10/sm_121a. 🔓 UNLOCKED (impossible until this release): ▸ DFlash + FP8 KV cache — 2× KV capacity, draft acceptance intact, measured on-device ▸ FP8 KV on vision models — the KV-bandwidth lever is back ▸ DFlash on FlashInfer & Triton backends (was FlashAttention-only) 🛠 Carried ahead of upstream (still-open PRs, merged + validated): ▸ Triton NVFP4 KV cache → up to 3× KV capacity ▸ DFlash sliding-window drafters + prefix-cache stability for DFlash ▸ High-concurrency DFlash — clean through c=64 🧰 Fixes stock v0.24.0 doesn't have: ▸ Gemma-4 tied embeddings (ModelOpt) — stock crashes at load, ours doesn't ▸ Tool calling out of the box (stock --no-deps builds 500) ▸ Unified-memory hardened — no silent KV overcommit on the 121 GB pool 📊 One image, the whole fleet: Gemma-4-26B-A4B · Qwen3.6-27B · Qwen3.6-35B-A3B (hybrid GDN) 536 tok/s @ c16 · 472 tok/s @ c12 · 262K context · image+video · tools + reasoning parsers · full CUDA-graph capture 🐳 docker pull

Hot off the press 🔥 AEON vLLM Ultimate → v0.24.0. Link below ⤵️ The ultimate container for AI on NVIDIA DGX Spark — vLLM compiled from source for GB10/sm_121a. 🔓 UNLOCKED (impossible until this release): ▸ DFlash + FP8 KV cache — 2× KV capacity, draft acceptance intact, measured on-device ▸ FP8 KV on vision models — the KV-bandwidth lever is back ▸ DFlash on FlashInfer & Triton backends (was FlashAttention-only) 🛠 Carried ahead of upstream (still-open PRs, merged + validated): ▸ Triton NVFP4 KV cache → up to 3× KV capacity ▸ DFlash sliding-window drafters + prefix-cache stability for DFlash ▸ High-concurrency DFlash — clean through c=64 🧰 Fixes stock v0.24.0 doesn't have: ▸ Gemma-4 tied embeddings (ModelOpt) — stock crashes at load, ours doesn't ▸ Tool calling out of the box (stock --no-deps builds 500) ▸ Unified-memory hardened — no silent KV overcommit on the 121 GB pool 📊 One image, the whole fleet: Gemma-4-26B-A4B · Qwen3.6-27B · Qwen3.6-35B-A3B (hybrid GDN) 536 tok/s @ c16 · 472 tok/s @ c12 · 262K context · image+video · tools + reasoning parsers · full CUDA-graph capture 🐳 docker pull

ÆON FORGE ✨

24,800 views • 3 months ago

NVIDIA is giving away free access to 140+ AI models for a full year > a lot of people building agents are paying $50-200/month just for API access NVIDIA just made that argument irrelevant models you get: GLM 5.2, MiniMax M3, Nemotron-3.5-lightning-30b-a3b, Muse-glimmer-30b from Meta and 130+ more setup: > step 1 - get your free key > go to > register -> bind phone -> copy API key > step 2 - add to Hermes agent > open Settings -> Model Provider -> Custom base_url = " api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx" > step 3 - pick a model model = "z-ai/glm-5.2" model = "minimaxai/minimax-m3" model = "nvidia/nemotron-3-ultra-550b-a55b" model = "moonshot-ai/kimi-k2.7" > Hermes already has NVIDIA set as default base_url > paste the key and you're running instantly > works the same in Cursor and OpenCode > cost: $0 > limit: 40 req/min > expires: 1 year a free key changes what you can afford to run: > one prompt costs nothing either way > a loop firing 300 agents every week is where the bill shows up > for a year that bill is zero while everyone is paying for API access, this is sitting there for free with this many models in one place it is also the easiest way to test which one actually fits your agent before you commit to a paid provider

NVIDIA is giving away free access to 140+ AI models for a full year > a lot of people building agents are paying $50-200/month just for API access NVIDIA just made that argument irrelevant models you get: GLM 5.2, MiniMax M3, Nemotron-3.5-lightning-30b-a3b, Muse-glimmer-30b from Meta and 130+ more setup: > step 1 - get your free key > go to > register -> bind phone -> copy API key > step 2 - add to Hermes agent > open Settings -> Model Provider -> Custom base_url = " api_key = "nvapi-xxxxxxxxxxxxxxxxxxxx" > step 3 - pick a model model = "z-ai/glm-5.2" model = "minimaxai/minimax-m3" model = "nvidia/nemotron-3-ultra-550b-a55b" model = "moonshot-ai/kimi-k2.7" > Hermes already has NVIDIA set as default base_url > paste the key and you're running instantly > works the same in Cursor and OpenCode > cost: $0 > limit: 40 req/min > expires: 1 year a free key changes what you can afford to run: > one prompt costs nothing either way > a loop firing 300 agents every week is where the bill shows up > for a year that bill is zero while everyone is paying for API access, this is sitting there for free with this many models in one place it is also the easiest way to test which one actually fits your agent before you commit to a paid provider

Annatar.md

208,474 views • 1 month ago

Nemotron 3 Full Breakdown With the help of Joey Conway from NVIDIA AI getting into the specifics around why Nemotron 3 is kind of a big deal Biggest headline with Nemotron is: Hybrid Mamba Transformer, Latent MoE, and MTP Hybrid Mamba Transformer essentially attacks right at the Attention mechanism to make the overhead sub-quadratic, but unlike quantizing KV Cache or swapping out attention head, NVIDIA chose Mamba-2 Latent MoE helps further optimize on sparsity by down projecting the dimensions so you're doing less math and less memory movement between HBM and SRAM, you're saving a ton, and NVIDIA made a conscious choice to add more experts given the surplus Finally, MTP or multi token prediction where the model can see future tokens to be more expressive in training and also option to use for speculative decoding during inference Oh, also the model adopts the new OpenMDW 1.1 License

Nemotron 3 Full Breakdown With the help of Joey Conway from NVIDIA AI getting into the specifics around why Nemotron 3 is kind of a big deal Biggest headline with Nemotron is: Hybrid Mamba Transformer, Latent MoE, and MTP Hybrid Mamba Transformer essentially attacks right at the Attention mechanism to make the overhead sub-quadratic, but unlike quantizing KV Cache or swapping out attention head, NVIDIA chose Mamba-2 Latent MoE helps further optimize on sparsity by down projecting the dimensions so you're doing less math and less memory movement between HBM and SRAM, you're saving a ton, and NVIDIA made a conscious choice to add more experts given the surplus Finally, MTP or multi token prediction where the model can see future tokens to be more expressive in training and also option to use for speculative decoding during inference Oh, also the model adopts the new OpenMDW 1.1 License

Caleb Eom

50,467 views • 4 months ago

We’re excited to share that 🥇Llama Nemotron Super 49B v1.5 -- our latest open reasoning model -- is now #1 on the Artificial Analysis Intelligence Index - a leaderboard that spans advanced math, science, and agentic tasks, in the 70B open model category. Llama Nemotron Super 49B v1.5 is trained with high-quality reasoning synthetic data generated from models like Qwen3-235B and DeepSeek R1. It delivers state-of-the-art accuracy and throughput, running on a single H100. Key features: 🎯 Leading accuracy on multi-step reasoning, math, coding, and function-calling 🏗️ Post-trained using RPO, DPO, and RLVR across 26M+ synthetic examples 📊 Fully transparent training data and techniques If you're building AI agents and want a high accuracy, fully-open, and transparent reasoning model that you can deploy anywhere, try Super v1.5 on or download from Hugging Face 🤗 ➡️ Leaderboard:

We’re excited to share that 🥇Llama Nemotron Super 49B v1.5 -- our latest open reasoning model -- is now #1 on the Artificial Analysis Intelligence Index - a leaderboard that spans advanced math, science, and agentic tasks, in the 70B open model category. Llama Nemotron Super 49B v1.5 is trained with high-quality reasoning synthetic data generated from models like Qwen3-235B and DeepSeek R1. It delivers state-of-the-art accuracy and throughput, running on a single H100. Key features: 🎯 Leading accuracy on multi-step reasoning, math, coding, and function-calling 🏗️ Post-trained using RPO, DPO, and RLVR across 26M+ synthetic examples 📊 Fully transparent training data and techniques If you're building AI agents and want a high accuracy, fully-open, and transparent reasoning model that you can deploy anywhere, try Super v1.5 on or download from Hugging Face 🤗 ➡️ Leaderboard:

NVIDIA AI Developer

100,506 views • 1 year ago

Most speculative decoding still drafts tokens one at a time. That's not parallel generation — it just hides the serial loop behind a smaller model. UC San Diego's z-lab just drew a clear line between the two. They released DFlash — a lightweight block diffusion model that drafts a whole block of tokens in a single forward pass, then lets the target model verify the block in parallel. Up to 15× higher throughput for gpt-oss-120b on NVIDIA Blackwell. No token-by-token drafting anywhere in the speculative path. Here's what's actually interesting: → The drafter is conditioned on the target model's own hidden features, injected into the Key/Value cache of every draft layer — so acceptance length scales with draft depth instead of diluting away → A 5-layer drafter replaces the 7B diffusion drafters that capped earlier methods near 3–4× → MATH-500 speedup: 6.08× vs. 1.81× for EAGLE-3 (4.86× average vs. 1.76×, Qwen3-8B, greedy) → Up to 15× higher throughput for gpt-oss-120b on NVIDIA Blackwell — at the same interactivity target → Lossless: the target still verifies every token, so output quality is preserved Full analysis: Paper: NVIDIA's metrics: Project: Model weights: Repo: NVIDIA AI NVIDIA AIDev

Most speculative decoding still drafts tokens one at a time. That's not parallel generation — it just hides the serial loop behind a smaller model. UC San Diego's z-lab just drew a clear line between the two. They released DFlash — a lightweight block diffusion model that drafts a whole block of tokens in a single forward pass, then lets the target model verify the block in parallel. Up to 15× higher throughput for gpt-oss-120b on NVIDIA Blackwell. No token-by-token drafting anywhere in the speculative path. Here's what's actually interesting: → The drafter is conditioned on the target model's own hidden features, injected into the Key/Value cache of every draft layer — so acceptance length scales with draft depth instead of diluting away → A 5-layer drafter replaces the 7B diffusion drafters that capped earlier methods near 3–4× → MATH-500 speedup: 6.08× vs. 1.81× for EAGLE-3 (4.86× average vs. 1.76×, Qwen3-8B, greedy) → Up to 15× higher throughput for gpt-oss-120b on NVIDIA Blackwell — at the same interactivity target → Lossless: the target still verifies every token, so output quality is preserved Full analysis: Paper: NVIDIA's metrics: Project: Model weights: Repo: NVIDIA AI NVIDIA AIDev

Marktechpost AI

23,328 views • 3 months ago

With Kimi K3 Day-0 on vLLM: Open Frontier Intelligence for Everyone 🚀 At 2.8 trillion parameters, Moonshot AI's Kimi K3 is one of the most powerful open-weight models ever released. Starting today, you can serve it on vLLM the moment the weights are public. What K3 brings: 🧠 2.8T-parameter Mixture-of-Experts (16 of 896 experts active per token) 📚 1M-token context window 🛠️ Native multimodal understanding, including vision ⚡ Kimi Delta Attention: a hybrid of linear and full attention that makes million-token context affordable Huge thank you to Kimi.ai AI for the model release and partnership, Inferact for leading the vLLM optimizations, and to our partners at NVIDIA, AMD, and the broader vLLM community. 1/6

With Kimi K3 Day-0 on vLLM: Open Frontier Intelligence for Everyone 🚀 At 2.8 trillion parameters, Moonshot AI's Kimi K3 is one of the most powerful open-weight models ever released. Starting today, you can serve it on vLLM the moment the weights are public. What K3 brings: 🧠 2.8T-parameter Mixture-of-Experts (16 of 896 experts active per token) 📚 1M-token context window 🛠️ Native multimodal understanding, including vision ⚡ Kimi Delta Attention: a hybrid of linear and full attention that makes million-token context affordable Huge thank you to Kimi.ai AI for the model release and partnership, Inferact for leading the vLLM optimizations, and to our partners at NVIDIA, AMD, and the broader vLLM community. 1/6

vLLM

104,533 views • 2 months ago

NVIDIA Releases Audex (Nemotron-Labs-Audex-30B-A3B): A Unified Audio-Text LLM That Preserves the Text Intelligence of Its Backbone Most unified audio models pay a text tax. Add audio output, and reasoning benchmarks drop — even when the only new output is speech. NVIDIA just released one that doesn't. Audex (Nemotron-Labs-Audex-30B-A3B) is a 30B MoE with 3B active parameters, built on the text-only Nemotron-Cascade-2-30B-A3B backbone. Audio inputs are projected into the text embedding space. Text tokens and quantized audio tokens are then generated the same way, inside one MoE decoder — no thinker–talker split, no stacked cascade. Here's what's actually interesting: → One model, audio in and out: understanding, ASR, translation, TTS, text-to-audio, speech-to-speech → Text holds vs its own backbone: IMO AnswerBench 81.1 vs 79.3, MMLU-Redux 86.4 vs 86.3 → Beats text-only Qwen3.5-35B-A3B on several tasks: LiveCodeBench v6 85.3 vs 74.6, IFBench 77.8 vs 70.2 → The usual tax, for contrast: Qwen3-Omni-30B-A3B-Thinking drops to 60.4 on HMMT vs 71.4 for its text backbone → 6.82 WER on OpenASR, ahead of Step-Audio-R1.1-33B (7.91) and Qwen3-Omni-Thinking (8.00) → Two codecs: X-Codec2 for speech (50 tok/s, FSQ, 65,536 codebook), X-Codec for general audio (200 tok/s, 4 flattened RVQ layers) — the only strong open model generating general audio beyond speech Full analysis: Paper: Model weights: NVIDIA AI NVIDIA Wei Ping

NVIDIA Releases Audex (Nemotron-Labs-Audex-30B-A3B): A Unified Audio-Text LLM That Preserves the Text Intelligence of Its Backbone Most unified audio models pay a text tax. Add audio output, and reasoning benchmarks drop — even when the only new output is speech. NVIDIA just released one that doesn't. Audex (Nemotron-Labs-Audex-30B-A3B) is a 30B MoE with 3B active parameters, built on the text-only Nemotron-Cascade-2-30B-A3B backbone. Audio inputs are projected into the text embedding space. Text tokens and quantized audio tokens are then generated the same way, inside one MoE decoder — no thinker–talker split, no stacked cascade. Here's what's actually interesting: → One model, audio in and out: understanding, ASR, translation, TTS, text-to-audio, speech-to-speech → Text holds vs its own backbone: IMO AnswerBench 81.1 vs 79.3, MMLU-Redux 86.4 vs 86.3 → Beats text-only Qwen3.5-35B-A3B on several tasks: LiveCodeBench v6 85.3 vs 74.6, IFBench 77.8 vs 70.2 → The usual tax, for contrast: Qwen3-Omni-30B-A3B-Thinking drops to 60.4 on HMMT vs 71.4 for its text backbone → 6.82 WER on OpenASR, ahead of Step-Audio-R1.1-33B (7.91) and Qwen3-Omni-Thinking (8.00) → Two codecs: X-Codec2 for speech (50 tok/s, FSQ, 65,536 codebook), X-Codec for general audio (200 tok/s, 4 flattened RVQ layers) — the only strong open model generating general audio beyond speech Full analysis: Paper: Model weights: NVIDIA AI NVIDIA Wei Ping

Marktechpost AI

28,301 views • 3 months ago

Nvidia has just announced Alpamayo 2 Super, an open 34 billion parameter reasoning vision-language-action model designed to accelerate the development of autonomous vehicles. This new model combines the NVIDIA Cosmos 3 Super reasoning model with a 2 billion parameter diffusion-based action expert model, and is post trained with reinforcement learning. The model can return multiple outputs: future trajectory plans, reasoning traces, grounded answers to questions about the scenes, and auto label generation. The model weights are now available for anyone to download on Hugging Face, and the inference code has been posted to GitHub. Distilled models can be deployed commercially without any further permission from Nvidia, and model outputs carry no license conditions. Automakers can distill down a compact version of this model that can run on the Nvidia computer in the car. Major kudos to Nvidia and Jensen Huang for advancing the state of the industry by releasing this as an open model with permissive licensing. Jensen isn't just paying lip service to the idea of open models, Nvidia is actually contributing to the ecosystem — and it's great for their business, because it helps sell more Thor computers that go in the car. Anyone can go download the model and play with it. If you do, let me know what you think. Personally I think it's so cool that we have open weights models that are this advanced, for anyone to download.

Nvidia has just announced Alpamayo 2 Super, an open 34 billion parameter reasoning vision-language-action model designed to accelerate the development of autonomous vehicles. This new model combines the NVIDIA Cosmos 3 Super reasoning model with a 2 billion parameter diffusion-based action expert model, and is post trained with reinforcement learning. The model can return multiple outputs: future trajectory plans, reasoning traces, grounded answers to questions about the scenes, and auto label generation. The model weights are now available for anyone to download on Hugging Face, and the inference code has been posted to GitHub. Distilled models can be deployed commercially without any further permission from Nvidia, and model outputs carry no license conditions. Automakers can distill down a compact version of this model that can run on the Nvidia computer in the car. Major kudos to Nvidia and Jensen Huang for advancing the state of the industry by releasing this as an open model with permissive licensing. Jensen isn't just paying lip service to the idea of open models, Nvidia is actually contributing to the ecosystem — and it's great for their business, because it helps sell more Thor computers that go in the car. Anyone can go download the model and play with it. If you do, let me know what you think. Personally I think it's so cool that we have open weights models that are this advanced, for anyone to download.

Whole Mars Catalog

45,595 views • 2 months ago

Introducing Open TTS Tracker! 🗣️ *sound on* A one-stop shop to track all open access/ source TTS models! Ranging from XTTS to Pheme, OpenVoice to VITS, and more... ⚡ For each model, we compile: 1. Souce-code 2. Checkpoints 3. License 4. Fine-tuning code 5. Languages supported 6. Paper 7. Demo Help us make it more complete! Let's 2024 the year of open TTS models! ❤️

Introducing Open TTS Tracker! 🗣️ sound on A one-stop shop to track all open access/ source TTS models! Ranging from XTTS to Pheme, OpenVoice to VITS, and more... ⚡ For each model, we compile: 1. Souce-code 2. Checkpoints 3. License 4. Fine-tuning code 5. Languages supported 6. Paper 7. Demo Help us make it more complete! Let's 2024 the year of open TTS models! ❤️

Vaibhav (VB) Srivastav

68,064 views • 2 years ago

Soofi Consortium Releases Soofi S 30B-A3B: An Open 31.6B Model for German and English Hitting 79.1 German Aggregate With Only 3.2B Active Parameters. Here's how it works. 👇 1. Sparsity in two places at once 52 layers: 23 Mamba-2, 23 granular MoE, 6 Grouped-Query Attention. The MoE router picks 6 of 128 experts per token, plus 2 shared. Mamba-2 carries the sequence mixing with a fixed-size recurrent state, so 46 of 52 layers keep no KV cache at all. → 3.2B of 31.6B parameters active per token 2. Reference architecture on purpose No bespoke backbone. It adopts NVIDIA's Nemotron 3 Nano design without modification — for day-one vLLM kernels, for serving efficiency, and for scientific control. That last one is the real move: Nemotron becomes an architecture-identical baseline, so the data recipe is the only variable left. 3. German as the deliberate variable Three-phase Warmup–Stable–Decay curriculum. Phase 1 is breadth at a 1e-3 plateau, Phase 2 concentrates high-quality data as the LR decays, Phase 3 stretches context to 1M tokens. → ~26.68T consumed tokens → German 7.2% → 15.32% of the mixture, vs ~5% for all non-English in the Nemotron reference → +4.2 German aggregate, +1.8 English, +6.7 held-out English over Nemotron 4. Where the architecture pays: memory bandwidth Every decoded token re-reads the weights and, for a Transformer, the attention cache of every sequence in the batch. Six KV layers instead of 52 keeps that per-sequence state small. Measured on one B200, TP=1, vLLM latency-subtraction. → 8–9× aggregate decode TPS/GPU vs dense 14–24B models at 40K context, batch 32 → decode stays flat from 4K to 256K 5. The numbers (base model, lm-evaluation-harness, 16 open baselines) → 70.1 English aggregate, +2.8 over Olmo 3 32B → 79.1 German aggregate, +6.3 over Apertus 70B → 73.8 HumanEval, 84.2 MBPP-DE, 88.8 GLP-DE, 61.2 INCLUDE-DE Full analysis: Paper: Technical details:

Soofi Consortium Releases Soofi S 30B-A3B: An Open 31.6B Model for German and English Hitting 79.1 German Aggregate With Only 3.2B Active Parameters. Here's how it works. 👇 1. Sparsity in two places at once 52 layers: 23 Mamba-2, 23 granular MoE, 6 Grouped-Query Attention. The MoE router picks 6 of 128 experts per token, plus 2 shared. Mamba-2 carries the sequence mixing with a fixed-size recurrent state, so 46 of 52 layers keep no KV cache at all. → 3.2B of 31.6B parameters active per token 2. Reference architecture on purpose No bespoke backbone. It adopts NVIDIA's Nemotron 3 Nano design without modification — for day-one vLLM kernels, for serving efficiency, and for scientific control. That last one is the real move: Nemotron becomes an architecture-identical baseline, so the data recipe is the only variable left. 3. German as the deliberate variable Three-phase Warmup–Stable–Decay curriculum. Phase 1 is breadth at a 1e-3 plateau, Phase 2 concentrates high-quality data as the LR decays, Phase 3 stretches context to 1M tokens. → ~26.68T consumed tokens → German 7.2% → 15.32% of the mixture, vs ~5% for all non-English in the Nemotron reference → +4.2 German aggregate, +1.8 English, +6.7 held-out English over Nemotron 4. Where the architecture pays: memory bandwidth Every decoded token re-reads the weights and, for a Transformer, the attention cache of every sequence in the batch. Six KV layers instead of 52 keeps that per-sequence state small. Measured on one B200, TP=1, vLLM latency-subtraction. → 8–9× aggregate decode TPS/GPU vs dense 14–24B models at 40K context, batch 32 → decode stays flat from 4K to 256K 5. The numbers (base model, lm-evaluation-harness, 16 open baselines) → 70.1 English aggregate, +2.8 over Olmo 3 32B → 79.1 German aggregate, +6.3 over Apertus 70B → 73.8 HumanEval, 84.2 MBPP-DE, 88.8 GLP-DE, 61.2 INCLUDE-DE Full analysis: Paper: Technical details:

Marktechpost AI

65,637 views • 2 months ago