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.

good morning /v1/chat/completions This is a test we ran overnight on TensorRT-LLM with modified kernels serving a custom 1B parameter model we trained for a customer ~200ms end-to-end latency (not TTFB, full request). Beats their current Cerebras stack on latency and quality

Sam Hogan 🇺🇸

22,674 subscribers

43,655 views • 6 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Really proud to share something we’ve been working on for a while: Magenta RealTime 2 (MTR2), a live music model that is highly interactive (MIDI, audio, text, lots of parameters) and low-latency (~200ms end-to-end), and runs locally on a MacBook!

Really proud to share something we’ve been working on for a while: Magenta RealTime 2 (MTR2), a live music model that is highly interactive (MIDI, audio, text, lots of parameters) and low-latency (~200ms end-to-end), and runs locally on a MacBook!

Ilaria Manco

45,072 views • 2 months ago

AI that responds in under 100ms doesn't happen by accident. Together AI VP of Kernels Dan Fu shares how they use NVIDIA CUDA, TensorRT-LLM, Dynamo and Together ATLAS on NVIDIA Blackwell to power ultra-low-latency inference and long context code generation for Cursor. From building megakernels that fuse an entire model's forward pass into a single launch to enabling real-time voice agents, Together AI is optimizing latency at every layer of the stack.

AI that responds in under 100ms doesn't happen by accident. Together AI VP of Kernels Dan Fu shares how they use NVIDIA CUDA, TensorRT-LLM, Dynamo and Together ATLAS on NVIDIA Blackwell to power ultra-low-latency inference and long context code generation for Cursor. From building megakernels that fuse an entire model's forward pass into a single launch to enabling real-time voice agents, Together AI is optimizing latency at every layer of the stack.

NVIDIA AI Infrastructure

20,633 views • 1 month ago

Multi-LoRA is in private preview on Cerebras Inference. Deploy one base model alongside a library of LoRA adapters. Switch between them per request, with no reloading, no separate deployments, and no latency cost. Available now for dedicated endpoint users. Reach out to your account rep to get access.

Multi-LoRA is in private preview on Cerebras Inference. Deploy one base model alongside a library of LoRA adapters. Switch between them per request, with no reloading, no separate deployments, and no latency cost. Available now for dedicated endpoint users. Reach out to your account rep to get access.

Cerebras

21,168 views • 2 months ago

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

kwindla

40,593 views • 2 months ago

"I used a billion tokens this week. I'm not even in the top 100 Codex users at OpenAI." We sat down with jason (creator of Instructor, now on OpenAI's Developer Experience team) to talk about how zero-latency inference is changing the way engineers work.

"I used a billion tokens this week. I'm not even in the top 100 Codex users at OpenAI." We sat down with jason (creator of Instructor, now on OpenAI's Developer Experience team) to talk about how zero-latency inference is changing the way engineers work.

Cerebras

93,023 views • 4 months ago

Inkling, Thinking Machines' first open model, dropped today: 975B total / 41B active MoE, up to 1M context, reasoning natively over text, images, and audio. Serving and RL support are already live: you can run and shape it on an open stack, starting now. Day 0 support on SGLang SGLang and Miles RadixArk👇 - Inkling's new architecture (ShortConv, attention with relative positional embedding, shared expert sink MoE) is natively implemented and deeply optimized, with prefill full CUDA graph and MXFP8 KV cache - Full parameter and LoRA RL in a customized Megatron backend, train inference consistency via customized kernels, routing replay, and cross-runtime parameter synchronization - DFlash speculative decoding from Modal for low-latency serving Launching now, blog and cookbook in the comments ⬇️

Inkling, Thinking Machines' first open model, dropped today: 975B total / 41B active MoE, up to 1M context, reasoning natively over text, images, and audio. Serving and RL support are already live: you can run and shape it on an open stack, starting now. Day 0 support on SGLang SGLang and Miles RadixArk👇 - Inkling's new architecture (ShortConv, attention with relative positional embedding, shared expert sink MoE) is natively implemented and deeply optimized, with prefill full CUDA graph and MXFP8 KV cache - Full parameter and LoRA RL in a customized Megatron backend, train inference consistency via customized kernels, routing replay, and cross-runtime parameter synchronization - DFlash speculative decoding from Modal for low-latency serving Launching now, blog and cookbook in the comments ⬇️

LMSYS Org

145,654 views • 1 month ago

Finally got to ~zero input-to-display latency on macOS. It has taken us a _lot_ of work to get here! This is a slow-motion capture of a 120 Hz screen running a playbit program that draws a rectangle where the cursor is. The slight lag you're seeing is not from rendering latency but from "old" mouse position data (we receive it from macOS window server in a event queue.) Remember that the mouse cursor is drawn with dedicated hardware (dedicated GPU plane) so dragging a rectangle is the ultimate input-to-display latency test.

Finally got to ~zero input-to-display latency on macOS. It has taken us a _lot_ of work to get here! This is a slow-motion capture of a 120 Hz screen running a playbit program that draws a rectangle where the cursor is. The slight lag you're seeing is not from rendering latency but from "old" mouse position data (we receive it from macOS window server in a event queue.) Remember that the mouse cursor is drawn with dedicated hardware (dedicated GPU plane) so dragging a rectangle is the ultimate input-to-display latency test.

Rasmus Andersson

214,759 views • 2 months ago

We work, we hold on, and we keep moving through a tunnel where no light is visible yet. And we know that darkness at the end of the tunnel means a turn, not the end. The most important thing is to keep going. Good will win.

We work, we hold on, and we keep moving through a tunnel where no light is visible yet. And we know that darkness at the end of the tunnel means a turn, not the end. The most important thing is to keep going. Good will win.

Anton Gerashchenko

70,798 views • 7 months ago

President Ruto: When we talk about pregnancy it’s not about women; no woman gets pregnant on their own. Men should not pretend that this is a women issue; in every pregnancy a man is involved. Additionally, childbirth doesn’t happen overnight; you have a 9 months’ notice. How we end up with an emergency on a situation we had 9 months’ notice is completely reckless

President Ruto: When we talk about pregnancy it’s not about women; no woman gets pregnant on their own. Men should not pretend that this is a women issue; in every pregnancy a man is involved. Additionally, childbirth doesn’t happen overnight; you have a 9 months’ notice. How we end up with an emergency on a situation we had 9 months’ notice is completely reckless

Citizen TV Kenya

97,220 views • 2 months ago

Newly released ✧ Gemma 4 is live on AI/ML API. We ran a benchmark test to showcase model capabilities on a complex analysis task, delivering stunning gains in speed, cost, and quality. RT + comment to win $100 API tokens - 50 random winners!

Newly released ✧ Gemma 4 is live on AI/ML API. We ran a benchmark test to showcase model capabilities on a complex analysis task, delivering stunning gains in speed, cost, and quality. RT + comment to win $100 API tokens - 50 random winners!

AI/ML API

26,455 views • 4 months ago

(1/5) FP4 hardware is here, but 4-bit attention still kills model quality, blocking true end-to-end FP4 serving. To fix that, we propose Attn-QAT, the first systematic study of quantization-aware training for attention. The result: FP4 attention quality is comparable to BF16 attention with 1.1x–1.5x higher throughput than SageAttention3 on an RTX 5090 and 1.39x speedup over FlashAttention-4 on a B200. Blog: Code: Checkpoints:

(1/5) FP4 hardware is here, but 4-bit attention still kills model quality, blocking true end-to-end FP4 serving. To fix that, we propose Attn-QAT, the first systematic study of quantization-aware training for attention. The result: FP4 attention quality is comparable to BF16 attention with 1.1x–1.5x higher throughput than SageAttention3 on an RTX 5090 and 1.39x speedup over FlashAttention-4 on a B200. Blog: Code: Checkpoints:

Hao AI Lab

37,869 views • 4 months ago

I trained a 12M parameter LLM on my own ML framework using a Rust backend and CUDA kernels for flash attention, AdamW, and more. Wrote the full transformer architecture, and BPE tokenizer from scratch. The framework features: - Custom CUDA kernels (Flash Attention, fused LayerNorm, fused GELU) for 3x increased throughput - Automatic WebGPU fallback for non-NVIDIA devices - TypeScript API with Rust compute backend - One npm install to get started, prebuilt binaries for every platform Try out the model for yourself: Built with Reese Chong. Check out the repos and blog if you want to learn more. Shoutout to Modal for the compute credits allowing me to train on 2 A100 GPUs without going broke cc sunny madra Gavin

I trained a 12M parameter LLM on my own ML framework using a Rust backend and CUDA kernels for flash attention, AdamW, and more. Wrote the full transformer architecture, and BPE tokenizer from scratch. The framework features: - Custom CUDA kernels (Flash Attention, fused LayerNorm, fused GELU) for 3x increased throughput - Automatic WebGPU fallback for non-NVIDIA devices - TypeScript API with Rust compute backend - One npm install to get started, prebuilt binaries for every platform Try out the model for yourself: Built with Reese Chong. Check out the repos and blog if you want to learn more. Shoutout to Modal for the compute credits allowing me to train on 2 A100 GPUs without going broke cc sunny madra Gavin

Aadi Kulshrestha

815,875 views • 4 months ago

🔥 This WEB3 GAME coming to APECHAIN is FIRE! 🔥 We teamed up with some content creators and gamers from Wayfinders to test CODE: Havoc , and it’s been a blast! 🔴 Here’s what stood out: - 3rd person shooter with fast-paced, fun gameplay - Tons of weapons - Runs super smooth—even on low-end PCs! (I had some latency since there are no LATAM servers yet ... TEST ) - Bet your ApeCoin in matches! 💰 This is definitely one to keep an eye on! 👀

🔥 This WEB3 GAME coming to APECHAIN is FIRE! 🔥 We teamed up with some content creators and gamers from Wayfinders to test CODE: Havoc , and it’s been a blast! 🔴 Here’s what stood out: - 3rd person shooter with fast-paced, fun gameplay - Tons of weapons - Runs super smooth—even on low-end PCs! (I had some latency since there are no LATAM servers yet ... TEST ) - Bet your ApeCoin in matches! 💰 This is definitely one to keep an eye on! 👀

EL REA 🇦🇷

10,813 views • 1 year ago

day 2 of building a self-driving power wheel today i officially trained a self driving model from scratch and deploy it on the car by just simply brute forcing everything, I: > made a remote tele-op and remote data collection app built on LiveKit infra > feat: 60ms e2e latency between the car and inference compute (car and compute in vietnam with singapore sfu) > feat: data is collected on operator side, baking latency into observation space itself (I expect this made the model more robust against latency) > recorded 30 min of data at 30fps and converted the dataset to lerobot (you can check a sample here) > trained a simple ACT model (3 epoch, batch size 8) to drive the car around my house > deployed the model on the car with remote inference the video explains everything shortly reflection: > the model is ofc bad, idt behavior cloning would work at all for such complex task on such small sample size > it did work in some cases where the observation is well within distribution, even generalizes to back the car when it gets stuck up next: > will hack alpamayo (NVIDIA) or comma ’s e2e to somehow fit this > or train with a llm backbone or a locomotion prior to see if it generalizes

day 2 of building a self-driving power wheel today i officially trained a self driving model from scratch and deploy it on the car by just simply brute forcing everything, I: > made a remote tele-op and remote data collection app built on LiveKit infra > feat: 60ms e2e latency between the car and inference compute (car and compute in vietnam with singapore sfu) > feat: data is collected on operator side, baking latency into observation space itself (I expect this made the model more robust against latency) > recorded 30 min of data at 30fps and converted the dataset to lerobot (you can check a sample here) > trained a simple ACT model (3 epoch, batch size 8) to drive the car around my house > deployed the model on the car with remote inference the video explains everything shortly reflection: > the model is ofc bad, idt behavior cloning would work at all for such complex task on such small sample size > it did work in some cases where the observation is well within distribution, even generalizes to back the car when it gets stuck up next: > will hack alpamayo (NVIDIA) or comma ’s e2e to somehow fit this > or train with a llm backbone or a locomotion prior to see if it generalizes

Binh Pham

19,404 views • 5 months ago

Today we’re releasing K2 Think V2, our most capable open-source reasoning model to date. This is a fully sovereign model: trained end-to-end on IFM-curated and synthesized data, with complete transparency from pre-training through final reasoning alignment.

Today we’re releasing K2 Think V2, our most capable open-source reasoning model to date. This is a fully sovereign model: trained end-to-end on IFM-curated and synthesized data, with complete transparency from pre-training through final reasoning alignment.

MBZUAI

287,728 views • 6 months ago

Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.

Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.

Freesolo

28,186 views • 1 month ago

The value of the work we're doing at Optimum is encapsulated quite well by the phrase "speed is money". In modern markets there are real economic advantages to latency reduction. This is nothing new. Wall Street firms have long been optimizing on latency, primarily through colocation and top of the line hardware. However, when it comes to decentralized systems, expensive hardware and geographic concentration are antithetical to their purpose. Therefore we should optimize decentralized network latency through software, which I'm thrilled about because it's exactly what I've spent the better part of the past 2 decades working on with Random Linear Network Coding. Now let’s talk about networking economics, the relationship between speed and money. First, it's important to note that users will only pay for low latency if it can be consistently guaranteed. Second, you can only make that latency guarantee for a certain number of users. This is a universal law of networking. We can model this relationship on a delay curve, shown below. The delay curve is determined by the utilization rate of the network, meaning how much traffic is flowing through the network divided by the network's throughput. As you approach a level of traffic equal to the available throughput, latency trends infinitely higher. On this delay curve we can impose some utility thresholds. These thresholds are the levels of latency which are important to different groups of users because of how that latency guarantee improves their economic outcomes. Finding the point on the curve where each threshold intersects will tell us what level of traffic we can guarantee that level of latency for. Essentially, there exists a finite supply of speed on a network and the highest utility users of that speed are willing to pay more for it. I like to think of this similarly to expedited shipping options on Amazon. This is why we say speed is money, and why we can create a Latency Marketplace. The only way to increase the supply of speed is to fundamentally increase network throughput. This is what we work on at Optimum by using Random Linear Network Coding. The same relationship between traffic and throughput still applies, but now the delay curve is shifted out further to the right. Now more traffic can be processed at the same latency, or the same traffic can be processed at a lower latency. More speed available to the network. More value unlocked for the network’s users. Crucially, that value is no longer only reserved for those who can afford to sit closest to the machine. Expanding the supply of speed widens who can reach each latency threshold, keeping the network's advantage decentralized rather than concentrated in the hands of a few. When nodes join Optimum and participate, they reap the benefits, but they also add to the capacity. Rather than vying against each other in a zero-sum game, nodes help themselves and others.

The value of the work we're doing at Optimum is encapsulated quite well by the phrase "speed is money". In modern markets there are real economic advantages to latency reduction. This is nothing new. Wall Street firms have long been optimizing on latency, primarily through colocation and top of the line hardware. However, when it comes to decentralized systems, expensive hardware and geographic concentration are antithetical to their purpose. Therefore we should optimize decentralized network latency through software, which I'm thrilled about because it's exactly what I've spent the better part of the past 2 decades working on with Random Linear Network Coding. Now let’s talk about networking economics, the relationship between speed and money. First, it's important to note that users will only pay for low latency if it can be consistently guaranteed. Second, you can only make that latency guarantee for a certain number of users. This is a universal law of networking. We can model this relationship on a delay curve, shown below. The delay curve is determined by the utilization rate of the network, meaning how much traffic is flowing through the network divided by the network's throughput. As you approach a level of traffic equal to the available throughput, latency trends infinitely higher. On this delay curve we can impose some utility thresholds. These thresholds are the levels of latency which are important to different groups of users because of how that latency guarantee improves their economic outcomes. Finding the point on the curve where each threshold intersects will tell us what level of traffic we can guarantee that level of latency for. Essentially, there exists a finite supply of speed on a network and the highest utility users of that speed are willing to pay more for it. I like to think of this similarly to expedited shipping options on Amazon. This is why we say speed is money, and why we can create a Latency Marketplace. The only way to increase the supply of speed is to fundamentally increase network throughput. This is what we work on at Optimum by using Random Linear Network Coding. The same relationship between traffic and throughput still applies, but now the delay curve is shifted out further to the right. Now more traffic can be processed at the same latency, or the same traffic can be processed at a lower latency. More speed available to the network. More value unlocked for the network’s users. Crucially, that value is no longer only reserved for those who can afford to sit closest to the machine. Expanding the supply of speed widens who can reach each latency threshold, keeping the network's advantage decentralized rather than concentrated in the hands of a few. When nodes join Optimum and participate, they reap the benefits, but they also add to the capacity. Rather than vying against each other in a zero-sum game, nodes help themselves and others.

Muriel Medard

44,395 views • 1 month ago

Next generation exchange Paradex is an onchain derivatives exchange built for serious traders. Built on Starknet (Privacy arc) 🥷, Paradex combines low-latency performance with full self-custody — CEX-level execution without giving up your keys. In this episode, we dive into how Paradex is re-architecting perps for the onchain era, their approach to scaling orderbooks on L2, and what a fully onchain trading stack could look like. Listen below:

Next generation exchange Paradex is an onchain derivatives exchange built for serious traders. Built on Starknet (Privacy arc) 🥷, Paradex combines low-latency performance with full self-custody — CEX-level execution without giving up your keys. In this episode, we dive into how Paradex is re-architecting perps for the onchain era, their approach to scaling orderbooks on L2, and what a fully onchain trading stack could look like. Listen below:

Jack Niewold 🫡

17,355 views • 9 months ago

🔊🔛🔥 ... Groq Inc launched voice generation today. GroqCloud now has realtime transcription, LLMs, *and* text-to-speech. You can build super-responsive, ultra low-latency voice agents end-to-end entirely on Groq! At Daily, we're big fans of Groq's fast, low-latency inference. Pipecat AI supports all the Groq models, including the new voice model. Lately, we've been obsessively playing a voice chat game that Mark Backman wrote. (My high score is 6, so far.) Here's Mark, with Groq's `Celeste-PlayAI` voice.

🔊🔛🔥 ... Groq Inc launched voice generation today. GroqCloud now has realtime transcription, LLMs, and text-to-speech. You can build super-responsive, ultra low-latency voice agents end-to-end entirely on Groq! At Daily, we're big fans of Groq's fast, low-latency inference. Pipecat AI supports all the Groq models, including the new voice model. Lately, we've been obsessively playing a voice chat game that Mark Backman wrote. (My high score is 6, so far.) Here's Mark, with Groq's `Celeste-PlayAI` voice.

kwindla

31,843 views • 1 year ago