正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

DeepSeek v4 Flash has 43 blocks... so, even 4096 batches during the prefill can be smoothly visualized. Look at the new progress bar.

antirez

73,333 subscribers

18,422 次观看 • 2 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

First steps for a specialized DeepSeek v4 Flash inference engine focused on inference quality / stability at different quantizations, with networked API that is batching capable. This is the 2 bit quants model running on my M3 Max 128GB.

First steps for a specialized DeepSeek v4 Flash inference engine focused on inference quality / stability at different quantizations, with networked API that is batching capable. This is the 2 bit quants model running on my M3 Max 128GB.

antirez

14,220 次观看 • 3 个月前

We ran a test between the new DeepSeek V4 Flash, Kimi K3 and GLM 5.2. 3 open models. 1 prompt. /design command Reviewed gameplay features, UX/UI and cost. 🔹 Kimi K3 → 9.5/10 · $0.0740 🔹 GLM 5.2 → 9/10 · $0.0480 🔹 DeepSeek V4 Flash → 7/10 · $0.0005 DeepSeek V4 Flash is ~150x cheaper, but rough around edges. Kimi K3 and GLM 5.2 built the most polished gaming experience.

We ran a test between the new DeepSeek V4 Flash, Kimi K3 and GLM 5.2. 3 open models. 1 prompt. /design command Reviewed gameplay features, UX/UI and cost. 🔹 Kimi K3 → 9.5/10 · $0.0740 🔹 GLM 5.2 → 9/10 · $0.0480 🔹 DeepSeek V4 Flash → 7/10 · $0.0005 DeepSeek V4 Flash is ~150x cheaper, but rough around edges. Kimi K3 and GLM 5.2 built the most polished gaming experience.

Command Code

165,918 次观看 • 11 天前

the way taehyung handled this so smoothly without even flinching, can never be me

the way taehyung handled this so smoothly without even flinching, can never be me

san

90,358 次观看 • 1 年前

Codex can now run Deepseek-v4- flash! There's a catch though. Deepseek's official setup switches your entire codex over to them, so your GPT models stop showing up at all. This is exactly what Codex Router is for. It adds models to the list instead of replacing them, so sol, grok, kimi and deepseek all sit in the same picker and i just grab whichever one suits the job. Deepseek v4-flash is $0.28 per million output tokens. opus 4.8 is $25. same picker, 89x apart. Links in the comment. setup's in the video 👇

Codex can now run Deepseek-v4- flash! There's a catch though. Deepseek's official setup switches your entire codex over to them, so your GPT models stop showing up at all. This is exactly what Codex Router is for. It adds models to the list instead of replacing them, so sol, grok, kimi and deepseek all sit in the same picker and i just grab whichever one suits the job. Deepseek v4-flash is $0.28 per million output tokens. opus 4.8 is $25. same picker, 89x apart. Links in the comment. setup's in the video 👇

Ziwen

145,018 次观看 • 11 天前

We just launched our 100% free coding agent everywhere. Every person in the world has access to 5 free hours of DeepSeek V4 Flash, everyday. npm i -g freebuff

We just launched our 100% free coding agent everywhere. Every person in the world has access to 5 free hours of DeepSeek V4 Flash, everyday. npm i -g freebuff

James Grugett

55,032 次观看 • 2 个月前

Has anyone noticed the mini Sylus on the progress bar? So cute! 😍

Has anyone noticed the mini Sylus on the progress bar? So cute! 😍

被ban的嘴替版⸆⸉

71,412 次观看 • 4 个月前

Can YRF ever come up with something original? Aditya Dhar has raised the bar so high that even the Khan gang can’t cross it.

Can YRF ever come up with something original? Aditya Dhar has raised the bar so high that even the Khan gang can’t cross it.

Moana

226,561 次观看 • 2 个月前

SwiftUI introduced a new modifier on iOS 26+ that essentially removes the thumb from the Slider and transforms it into a draggable progress bar. Currently, there’s no way to adjust the size of the progress bar. However, by using the scale and other modifiers, we can create a progress bar that is similar to the iOS Volume slider! Modifier: sliderThumbVisibility()

SwiftUI introduced a new modifier on iOS 26+ that essentially removes the thumb from the Slider and transforms it into a draggable progress bar. Currently, there’s no way to adjust the size of the progress bar. However, by using the scale and other modifiers, we can create a progress bar that is similar to the iOS Volume slider! Modifier: sliderThumbVisibility()

Kavsoft

22,783 次观看 • 9 个月前

Happy Friday! We just put DeepSeek-V4-Pro up on It’s the world’s largest open source model at 1.6T parameters, and you can run it for free running on NVIDIA Blackwell GPUs. Try the NVIDIA NIM API →

Happy Friday! We just put DeepSeek-V4-Pro up on It’s the world’s largest open source model at 1.6T parameters, and you can run it for free running on NVIDIA Blackwell GPUs. Try the NVIDIA NIM API →

NVIDIA AI

205,507 次观看 • 3 个月前

🔒 Introducing the HT V4 with Binding Loop—Next-Level Control Awaits! 🔒 A new accessory for the HT V4!! This ring has a non-removable steel ring attached to its top, which when affixed to the HT V4 chastity cage allows a lead to be attached, and the wearer to be led around. Endless possibilities ! 🤩

Sensitive content

🔒 Introducing the HT V4 with Binding Loop—Next-Level Control Awaits! 🔒 A new accessory for the HT V4!! This ring has a non-removable steel ring attached to its top, which when affixed to the HT V4 chastity cage allows a lead to be attached, and the wearer to be led around. Endless possibilities ! 🤩

House Of Chastity

28,442 次观看 • 1 年前

DeepSeek V4 Flash 0731 is now 90% off on Nous Portal for the next 7 days, in partnership with Novita AI. At this discounted price, it is over 1000x cheaper than Fable 5 on comparable tasks while still beating it on Terminal-Bench 2.1. Try it today at

DeepSeek V4 Flash 0731 is now 90% off on Nous Portal for the next 7 days, in partnership with Novita AI. At this discounted price, it is over 1000x cheaper than Fable 5 on comparable tasks while still beating it on Terminal-Bench 2.1. Try it today at

Nous Research

847,372 次观看 • 9 天前

Oisín Mullin's speed here is ridiculous. Look at him tearing away from Brisbane star Lachie Neale at the start. He has become the AFL's best tagger and is now showing he can be even more.

Oisín Mullin's speed here is ridiculous. Look at him tearing away from Brisbane star Lachie Neale at the start. He has become the AFL's best tagger and is now showing he can be even more.

Maurice Brosnan

182,421 次观看 • 2 个月前

look at this diva’s ability to stay in the center even during chaos

look at this diva’s ability to stay in the center even during chaos

jungkook admirer

534,949 次观看 • 25 天前

Saw a lot of dumb posts talking about how the new avatar fighting game "looks like a flash game" decide to look at the actual flash game people are comparing it to ...yeah i can throw those opinions in the dumpster lmao

Rooflemonger 🦍🦧🐵

1,974,712 次观看 • 9 个月前

Being a part of Alliance means I get the best gear a gamer can own. Unpacking the new Razer Rogue V4 Backpack so I can show off at LAN to everyone else 😜 R Λ Z Ξ R #ForThePro

Being a part of Alliance means I get the best gear a gamer can own. Unpacking the new Razer Rogue V4 Backpack so I can show off at LAN to everyone else 😜 R Λ Z Ξ R #ForThePro

Alliance akku

11,811 次观看 • 1 个月前

🚨 A New Look at the Dilophosaurus in 'JURASSIC WORLD REBIRTH' has been revealed! 🚨 This is our best look at the dinosaur so far, and it looks fantastic! 🦖

🚨 A New Look at the Dilophosaurus in 'JURASSIC WORLD REBIRTH' has been revealed! 🚨 This is our best look at the dinosaur so far, and it looks fantastic! 🦖

Swrve 🦖 'Jurassic' News & Content

32,988 次观看 • 1 年前

I inspected progress of construction at the South C property where the Law Society of Kenya secretariat will be relocated to during the construction of wakili Towers. The finishing is ongoing and the new offices should be ready for occupancy soon.

I inspected progress of construction at the South C property where the Law Society of Kenya secretariat will be relocated to during the construction of wakili Towers. The finishing is ongoing and the new offices should be ready for occupancy soon.

Faith Odhiambo

57,348 次观看 • 6 个月前

Day 12/90 of Inference Engineering What is chunked prefill within vLLM? In continuation of yesterday's post on the high level architecture of vLLM, I want to dive deeper into vLLM core engine starting with the mechanics of chunked prefill. In this post, I will closely follow the original blog on the anatomy of vLLM. To start, let's define chunked prefill. It's a runtime inference optimization technique that splits a long input request so that it doesn’t monopolize the whole GPU. Keep in mind this is all within the context of vLLM. And since vLLM is an inference engine that's meant to serve a model to multiple concurrent users, having a GPU that’s fully monopolized on a single user's request means other users' requests would be in queue waiting to be processed. It isn’t too good to have the whole GPU occupied on a single request when the GPU is meant to be shared! So the key idea behind chunked prefill is to break the long request into smaller chunks, so that each chunk along with other users' requests gets processed and written into the KV cache together. Suppose we split up the long request into chunks and each chunk has 8 tokens. Now each memory block can hold 4 tokens. Therefore, 8 tokens can fit into 2 blocks of memory. After the first forward pass, 2 blocks are occupied, and after the second forward pass, 4 blocks of memory are occupied and so forth. Each forward pass handles a small chunk of the long request so that there's room in the same pass to keep serving other users' requests. Here's a small animation that I made today to fully visualize the idea behind chunked prefill when learning this topic~

Day 12/90 of Inference Engineering What is chunked prefill within vLLM? In continuation of yesterday's post on the high level architecture of vLLM, I want to dive deeper into vLLM core engine starting with the mechanics of chunked prefill. In this post, I will closely follow the original blog on the anatomy of vLLM. To start, let's define chunked prefill. It's a runtime inference optimization technique that splits a long input request so that it doesn’t monopolize the whole GPU. Keep in mind this is all within the context of vLLM. And since vLLM is an inference engine that's meant to serve a model to multiple concurrent users, having a GPU that’s fully monopolized on a single user's request means other users' requests would be in queue waiting to be processed. It isn’t too good to have the whole GPU occupied on a single request when the GPU is meant to be shared! So the key idea behind chunked prefill is to break the long request into smaller chunks, so that each chunk along with other users' requests gets processed and written into the KV cache together. Suppose we split up the long request into chunks and each chunk has 8 tokens. Now each memory block can hold 4 tokens. Therefore, 8 tokens can fit into 2 blocks of memory. After the first forward pass, 2 blocks are occupied, and after the second forward pass, 4 blocks of memory are occupied and so forth. Each forward pass handles a small chunk of the long request so that there's room in the same pass to keep serving other users' requests. Here's a small animation that I made today to fully visualize the idea behind chunked prefill when learning this topic~

max fu

29,197 次观看 • 24 天前

Here's a look at the zombies the new hero Graves can spawn

Here's a look at the zombies the new hero Graves can spawn

Deadlock Intel

69,401 次观看 • 6 个月前

NEW: Tom Morrow 2.0 will be the host of the Walt Disney Imagineering pavilion at D23: The Ultimate Disney Fan Event! The pavilion will showcase the progress shaping Disney Parks worldwide and give fans an up-close look at what’s to come.

NEW: Tom Morrow 2.0 will be the host of the Walt Disney Imagineering pavilion at D23: The Ultimate Disney Fan Event! The pavilion will showcase the progress shaping Disney Parks worldwide and give fans an up-close look at what’s to come.

Drew Smith

20,491 次观看 • 15 天前