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.

1/ we just publicly released Muse Spark 1.3 max! we see significantly stronger coding and agentic performance on muse spark 1.3 max, so would strongly recommend trying it out even if you've already tried muse spark 1.3 high or muse spark 1.3 xhigh.

Alexandr Wang

654,668 subscribers

211,367 views • 15 hours ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

holy sh*t. this is f**king insane. muse spark 1.3 beats fable 5 and gpt 5.6 at 1/20th price i cancelled my $200/mo Claude subscription for this. i replaced fable 5 with muse spark 1.3 for only $10/mo on opencode go [it takes 3 minutes to set up:] 1/ install the model-router repo 2/ toggle opencode free to green 3/ that’s it.

holy sh*t. this is f**king insane. muse spark 1.3 beats fable 5 and gpt 5.6 at 1/20th price i cancelled my $200/mo Claude subscription for this. i replaced fable 5 with muse spark 1.3 for only $10/mo on opencode go [it takes 3 minutes to set up:] 1/ install the model-router repo 2/ toggle opencode free to green 3/ that’s it.

Avid

63,410 views • 1 day ago

1/ muse spark 1.2 is a very strong multimodal model—it can do visual coding, robotics planning, and audio-visual understanding that all come together through agentic tools.

1/ muse spark 1.2 is a very strong multimodal model—it can do visual coding, robotics planning, and audio-visual understanding that all come together through agentic tools.

Alexandr Wang

163,736 views • 15 days ago

Introducing Muse Image and Muse Video, the first media generation models developed by Meta Superintelligence Labs. Muse Image is our most advanced image generation model yet. It follows instructions faithfully, edits with precision, composes from multiple references, and draws on Instagram for social context. It also brings agentic tool use capabilities to image generation and integrates with Muse Spark. You can try Muse Image in the Meta AI app and web, as well as in Instagram Stories and WhatsApp – starting in limited countries with more locations on the way. Today we’re also previewing Muse Video, which is built upon the same pretraining base as Muse Image to deliver exceptional visual fidelity with native audio support. Learn more about both models:

Introducing Muse Image and Muse Video, the first media generation models developed by Meta Superintelligence Labs. Muse Image is our most advanced image generation model yet. It follows instructions faithfully, edits with precision, composes from multiple references, and draws on Instagram for social context. It also brings agentic tool use capabilities to image generation and integrates with Muse Spark. You can try Muse Image in the Meta AI app and web, as well as in Instagram Stories and WhatsApp – starting in limited countries with more locations on the way. Today we’re also previewing Muse Video, which is built upon the same pretraining base as Muse Image to deliver exceptional visual fidelity with native audio support. Learn more about both models:

AI at Meta

850,438 views • 1 month ago

Zuckerberg is serial copycat… he copies Elon at every turn: • Threads - an X/Twitter clone launched in July 2023 • Cage fight - he challenged him, hyped it up, then backed down like a bitch • Agentic AI & coding - now dropping Muse Spark 1.1 with the exact same “agentic, tool-using, multi-agent coding” features xAI/Grok has been pioneering • Posting on 𝕏 - after 3 full years of silence, posts on the world’s town square bc it’s simply the best option Build something original for once, Zuck.

Zuckerberg is serial copycat… he copies Elon at every turn: • Threads - an X/Twitter clone launched in July 2023 • Cage fight - he challenged him, hyped it up, then backed down like a bitch • Agentic AI & coding - now dropping Muse Spark 1.1 with the exact same “agentic, tool-using, multi-agent coding” features xAI/Grok has been pioneering • Posting on 𝕏 - after 3 full years of silence, posts on the world’s town square bc it’s simply the best option Build something original for once, Zuck.

Teslaconomics

69,959 views • 1 month ago

Grok 4.5 performed GPT Sol level for free! We gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics demos Prompts: -robot deathmatch, Tombstone vs Minotaur -a hydraulic press flattening stuff on a conveyor -a semi truck jumping a canyon Outputs: GPT-5.6 Sol: 12.9K tokens, $0.51 (~7 min) Grok 4.5: 10.8K tokens, $0 (~5 min) Muse Spark 1.1: 26.8K tokens, $0.12 (~7.5 min) GLM 5.2: 10.9K tokens, $0.02 (~12 min) Grok 4.5 handled all three scenes genuinely well and got surprisingly close to GPT-5.6 this round. On top of that, it ran on the free tier. GPT-5.6 Sol, the frontier model, put out solid but not standout work. GLM 5.2 rendered all three scenes for pennies, but it came out the roughest of the four. Meta's new Muse Spark burned the most tokens yet still stayed cheap, delivering an average result.

Grok 4.5 performed GPT Sol level for free! We gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics demos Prompts: -robot deathmatch, Tombstone vs Minotaur -a hydraulic press flattening stuff on a conveyor -a semi truck jumping a canyon Outputs: GPT-5.6 Sol: 12.9K tokens, $0.51 (~7 min) Grok 4.5: 10.8K tokens, $0 (~5 min) Muse Spark 1.1: 26.8K tokens, $0.12 (~7.5 min) GLM 5.2: 10.9K tokens, $0.02 (~12 min) Grok 4.5 handled all three scenes genuinely well and got surprisingly close to GPT-5.6 this round. On top of that, it ran on the free tier. GPT-5.6 Sol, the frontier model, put out solid but not standout work. GLM 5.2 rendered all three scenes for pennies, but it came out the roughest of the four. Meta's new Muse Spark burned the most tokens yet still stayed cheap, delivering an average result.

atomic.chat

70,490 views • 1 month ago

It wouldn’t be a #ChristianSiriano show without all the drama—especially with his muse, #CocoRocha, giving us the high fashion glamour we so desperately needed. Christian Siriano kicks off #NYFW and the entire Fashion Month on a high note. See more of the show in our stories.

It wouldn’t be a #ChristianSiriano show without all the drama—especially with his muse, #CocoRocha, giving us the high fashion glamour we so desperately needed. Christian Siriano kicks off #NYFW and the entire Fashion Month on a high note. See more of the show in our stories.

Fashion Week

22,467 views • 3 years ago

Meta just launched its 1st image model after Mark Zuckerberg’s AI shake-up. Muse Image is Meta Superintelligence Labs' first image generator after Muse Spark. They said the model will power new editing features in Meta’s Instagram photo app and be added to its marketer tools for creating platform ads. Meta had relied on Midjourney and Black Forest Labs for generation inside Meta AI. Muse Image brings that layer in-house, so Meta controls quality, cost, and product timing. Consumers get free access through Meta AI, WhatsApp chats, and Instagram Stories. Power users need Meta One or another monthly plan when free limits run out. Users can start from prompts, add photos, annotate edits, or sketch changes directly. Meta says it can erase photobombers, create QR codes, and keep visual text readable. Advertisers get variants through Advantage+ creative, with edits, style swaps, and brand-matched versions. Meta says internal tests trail GPT Image 2 but beat Nano Banana 2 on editing. This is another move in Meta's effort in trying to convert AI infrastructure spending into revenue beyond social ads.

Rohan Paul

20,377 views • 1 month ago

If the Universe’s only way to realize itself is through intelligent beings, why did it not protect life during earth’s evolution? Or does it have other purposes — and are we merely a byproduct of its creation? : If the Universe can only know itself through conscious minds, why did it let life on Earth teeter on the edge of extinction dozens of times? Asteroids, ice ages, supervolcanoes, and mass die-offs all struck without mercy. No shield, no mercy, no intervention. That silence is the answer. The Universe is not a guardian. It is an arena. It does not care whether intelligence ever arises, survives, or dies. The emergence of minds is not its goal; it is an accidental side effect of laws that permit complexity. We are not protected children of a loving cosmos. We are the rare, fragile spark that sometimes flickers into being inside an indifferent storm. Our existence proves nothing about cosmic purpose. It only proves that, out of countless dead rocks, one managed to keep the spark alive long enough to ask why. The Universe has no obligation to us. We have an obligation to the spark. That is the whole story.

If the Universe’s only way to realize itself is through intelligent beings, why did it not protect life during earth’s evolution? Or does it have other purposes — and are we merely a byproduct of its creation? : If the Universe can only know itself through conscious minds, why did it let life on Earth teeter on the edge of extinction dozens of times? Asteroids, ice ages, supervolcanoes, and mass die-offs all struck without mercy. No shield, no mercy, no intervention. That silence is the answer. The Universe is not a guardian. It is an arena. It does not care whether intelligence ever arises, survives, or dies. The emergence of minds is not its goal; it is an accidental side effect of laws that permit complexity. We are not protected children of a loving cosmos. We are the rare, fragile spark that sometimes flickers into being inside an indifferent storm. Our existence proves nothing about cosmic purpose. It only proves that, out of countless dead rocks, one managed to keep the spark alive long enough to ask why. The Universe has no obligation to us. We have an obligation to the spark. That is the whole story.

Zafar Mirzo | Quotes

272,822 views • 9 months ago

Muse Glimmer, A 30B parameter dense model swallowing a 130,000 token context window using only 19.3 GB of VRAM (extreme efficiency). No KV cache quantization required. I just benched the new Muse Glimmer 30B (dense) on a single RTX 4090. We are pulling 3,100+ t/s prefill and 75 tokens/second decode. The throughput is violent. Meta superintelligence lab just open sourced this agentic beast, explicitly engineered to dominate 24GB consumer cards. I pulled the latest llama.cpp source on Ubuntu 22 (CUDA 13) to see if the specs were real. Fed it a 28k token prompt. Here is the exact llama.cpp God Stack and benchmarking breakdown: # 1. The Deep Context Run (No Speculative Decoding) The architecture uses a massive 16:1 GQA (Grouped Query Attention) ratio. This means the KV cache footprint is practically non existent. ./build/bin/llama-server -m Muse-Glimmer-30B-UD-Q4_K_XL.gguf -c 130000 -b 4096 -ub 4096 -ngl 99 --port 8080 Prefill: 3134.95 t/s Decode: 50.00 t/s VRAM: 19.34 GB (I hit 130k context on pristine, unquantized f16 cache and still had 4.5 GB of VRAM left over. Absolute witchcraft). # 2. The DFlash Speculative Overdrive Meta shipped this with a DFlash block diffusion drafter. Let's trade that extra VRAM for pure speed. ./build/bin/llama-server -m Muse-Glimmer-30B-UD-Q4_K_XL.gguf -md dflash-kquant.gguf --spec-type draft-dflash --spec-draft-n-max 3 -c 80000 -b 4096 -ub 4096 -ngl 99 --port 8080 Prefill: 1293.69 t/s Decode: 75.00 t/s VRAM: 23.93 GB (Maxed out on card) the dflash gguf is additional 1.6 GBs # The Architecture Insight (Muse Glimmer vs. Gemma 4 31B) If you look at my Gemma 4 31B tests from last week, getting 140k context required heavily degrading the memory with Q4 KV quantization (gemma 31b q4 can do only about 40k context with unquantized kv on a 24gb card). That "unzipping" overhead bottlenecked Gemma's MTP decode speeds down to 65 t/s. Muse Glimmer completely sidesteps this bottleneck. By using aggressive 16:1 GQA, it keeps the KV cache in native f16 format at massive context lengths. Flash Attention gets to run at maximum uncompressed speed, letting the DFlash drafter push decode safely to 75 t/s without compute lag. With a 76% on SWE Bench Verified and seamless local tool calling, this model looks promising. Unsloth's Hugging Face GGUF links, intelligence/agentic benchmark details, and inference throughput performance graphs are posted in the replies. For 24GB rig, what’s your current go to model?

Muse Glimmer, A 30B parameter dense model swallowing a 130,000 token context window using only 19.3 GB of VRAM (extreme efficiency). No KV cache quantization required. I just benched the new Muse Glimmer 30B (dense) on a single RTX 4090. We are pulling 3,100+ t/s prefill and 75 tokens/second decode. The throughput is violent. Meta superintelligence lab just open sourced this agentic beast, explicitly engineered to dominate 24GB consumer cards. I pulled the latest llama.cpp source on Ubuntu 22 (CUDA 13) to see if the specs were real. Fed it a 28k token prompt. Here is the exact llama.cpp God Stack and benchmarking breakdown: # 1. The Deep Context Run (No Speculative Decoding) The architecture uses a massive 16:1 GQA (Grouped Query Attention) ratio. This means the KV cache footprint is practically non existent. ./build/bin/llama-server -m Muse-Glimmer-30B-UD-Q4_K_XL.gguf -c 130000 -b 4096 -ub 4096 -ngl 99 --port 8080 Prefill: 3134.95 t/s Decode: 50.00 t/s VRAM: 19.34 GB (I hit 130k context on pristine, unquantized f16 cache and still had 4.5 GB of VRAM left over. Absolute witchcraft). # 2. The DFlash Speculative Overdrive Meta shipped this with a DFlash block diffusion drafter. Let's trade that extra VRAM for pure speed. ./build/bin/llama-server -m Muse-Glimmer-30B-UD-Q4_K_XL.gguf -md dflash-kquant.gguf --spec-type draft-dflash --spec-draft-n-max 3 -c 80000 -b 4096 -ub 4096 -ngl 99 --port 8080 Prefill: 1293.69 t/s Decode: 75.00 t/s VRAM: 23.93 GB (Maxed out on card) the dflash gguf is additional 1.6 GBs # The Architecture Insight (Muse Glimmer vs. Gemma 4 31B) If you look at my Gemma 4 31B tests from last week, getting 140k context required heavily degrading the memory with Q4 KV quantization (gemma 31b q4 can do only about 40k context with unquantized kv on a 24gb card). That "unzipping" overhead bottlenecked Gemma's MTP decode speeds down to 65 t/s. Muse Glimmer completely sidesteps this bottleneck. By using aggressive 16:1 GQA, it keeps the KV cache in native f16 format at massive context lengths. Flash Attention gets to run at maximum uncompressed speed, letting the DFlash drafter push decode safely to 75 t/s without compute lag. With a 76% on SWE Bench Verified and seamless local tool calling, this model looks promising. Unsloth's Hugging Face GGUF links, intelligence/agentic benchmark details, and inference throughput performance graphs are posted in the replies. For 24GB rig, what’s your current go to model?

Alok

65,480 views • 25 days ago

Since Ryan Zeferjahn joined the Cubs, he's lost 1.3" of vertical break on his 4-Seamer, and slug against it has risen by .433 in August compared to last month. Why is this happening? No idea! No noticeable difference in grip, release position, velocity, or even raw spin rate. I wonder if the differences in barometric pressure, air density, or elevation between Angel Stadium and Wrigley are hurting the shape (pure guess, would love to hear others' thoughts). Or it's just natural fluctuation of the shape and this is an overreaction to small samples :)

Since Ryan Zeferjahn joined the Cubs, he's lost 1.3" of vertical break on his 4-Seamer, and slug against it has risen by .433 in August compared to last month. Why is this happening? No idea! No noticeable difference in grip, release position, velocity, or even raw spin rate. I wonder if the differences in barometric pressure, air density, or elevation between Angel Stadium and Wrigley are hurting the shape (pure guess, would love to hear others' thoughts). Or it's just natural fluctuation of the shape and this is an overreaction to small samples :)

Nate Roper

68,708 views • 17 days ago

Kree8 website will NOT lag now! Just ditched Framer and we'll be coding our future client websites - glad we took another risk of trying something new No-code tools are great… until you care about performance. I was so shocked to see how our website worked on Safari, and I have a MacBook Pro (M4 Max) 💀 Switched to code (Next.js) and suddenly everything feels instant. 👇With Framer With NextJS👇

Kree8 website will NOT lag now! Just ditched Framer and we'll be coding our future client websites - glad we took another risk of trying something new No-code tools are great… until you care about performance. I was so shocked to see how our website worked on Safari, and I have a MacBook Pro (M4 Max) 💀 Switched to code (Next.js) and suddenly everything feels instant. 👇With Framer With NextJS👇

Jay Dwivedi

59,684 views • 4 months ago

$🚨This is exactly what I was warning y'all about... $BTC is about to restart the entire cycle Accumulation phase is almost done But there are 2 things left to blow that chart UP If you still don't understand what's about to happen... Read this for the sake of your port.: The US Treasury just doubled its bond buyback size. $2B to $4B per operation. starting Sept 9. Analysts are calling it what it actually is. mini QE. That headline today already pushed yields down and sent BTC through $69K in hours. (it hasn't even started yet) Heres thing two: $41M got liquidated in the last 2 days. 96% of it was shorts getting steamrolled. (+$1.1B liquidated in 1h lol) Thats a fraction of the short interest piled up over the last month. (most of them are still trapped) We already got the spark AND the full tank of fuel. And the most important spark to light things UP is Clarity Act If it clears the Senate: - Every mechanic above gets amplified by trillions in institutional money finally let off the sidelines. I was screaming about it month ago... Notifs on bro.$

🚨This is exactly what I was warning y'all about... $BTC is about to restart the entire cycle Accumulation phase is almost done But there are 2 things left to blow that chart UP If you still don't understand what's about to happen... Read this for the sake of your port.: The US Treasury just doubled its bond buyback size. $2B to $4B per operation. starting Sept 9. Analysts are calling it what it actually is. mini QE. That headline today already pushed yields down and sent BTC through $69K in hours. (it hasn't even started yet) Heres thing two: $41M got liquidated in the last 2 days. 96% of it was shorts getting steamrolled. (+$1.1B liquidated in 1h lol) Thats a fraction of the short interest piled up over the last month. (most of them are still trapped) We already got the spark AND the full tank of fuel. And the most important spark to light things UP is Clarity Act If it clears the Senate: - Every mechanic above gets amplified by trillions in institutional money finally let off the sidelines. I was screaming about it month ago... Notifs on bro.

Frogify

20,366 views • 16 days ago

Qwen 3.8 27B Q4_K_M - 90 tokens/sec on a single NVIDIA RTX 4090 (24 GB VRAM) with Dflash2! (MTP 60 tps -> 90 tps Dflash2!!!!) Local AI moves so fast (literally!) it’s terrifying. Z lab just dropped DFlash 2 for Qwen 3.8 27b and Muse Glimmer. I patched llama.cpp (PR #27342) and paired it with Unsloth’s Qwen 3.8 27B UD-Q4_K_XL quant. The result? Lossless 90 tokens/s decode. My last post highlighted native MTP hitting 60 t/s at 130,000 context. But DFlash 2 just completely shattered that ceiling. By using parallel block diffusion drafting (predicting whole blocks of tokens in a single pass using dynamic convolutions), DFlash achieves a massive 5.39 token acceptance rate. THE ALPHA TWEAK: `n-max 7` eats too much VRAM for draft states. But if you drop the draft limit to `--spec-draft-n-max 4`, you slash the VRAM overhead and actually increase the throughput. Here is the new 24GB VRAM Physics Matrix (DFlash 2 @ n-max 4): - 30k Context: 1,725 t/s prefill | 87.05 t/s decode | 22.2 GB VRAM - 80k Context: 1,789 t/s prefill | 84.20 t/s decode | 23.3 GB VRAM - 110k Context: 1,767 t/s prefill | 83.35 t/s decode | 23.96 GB VRAM (110k context at 83+ tokens a second sitting exactly on the 24GB hardware limit is absolute wizardry). How to compile the PR today: git clone cd llama.cpp git fetch origin pull/27342/head:pr-27342 git switch pr-27342 cmake -B build -DGGML_CUDA=ON && cmake --build build -j Llama.cpp flags for Dflash (110k Context Ceiling): ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -md Qwen3.8-27B-DFlash2-Q4_K_M.gguf --spec-type draft-dflash --spec-draft-n-max 4 -c 110000 -ngl 99 --port 8080 -ctv q4_0 -ctk q4_0 The fact that the open source community is shipping block diffusion drafters so quickly that run entirely locally on a gaming GPU is unbelievable. If you own a single RTX 3090 or 4090, it is officially time to upgrade to qwen 3.8 27b with dflash 2 and cancel your API subscriptions and let local silicon eat the cloud. This model beats GPT 5.6 Terra, GLM 5.2 DeepSeek V4 Pro, Muse Spark 1.2 and Claude Opus 4.8 on the artificial analysis agentic index (details in the replies) Hugging Face GGUF links (Base + DFlash2) and the full visual VRAM scaling and Dflash2 vs MTP graphs are also in the replies below. are you sticking to native MTP for the 130k context, or sacrificing 20k context to redline your decode speed? How many tokens/sec are you pushing on your current local rig?

Qwen 3.8 27B Q4_K_M - 90 tokens/sec on a single NVIDIA RTX 4090 (24 GB VRAM) with Dflash2! (MTP 60 tps -> 90 tps Dflash2!!!!) Local AI moves so fast (literally!) it’s terrifying. Z lab just dropped DFlash 2 for Qwen 3.8 27b and Muse Glimmer. I patched llama.cpp (PR #27342) and paired it with Unsloth’s Qwen 3.8 27B UD-Q4_K_XL quant. The result? Lossless 90 tokens/s decode. My last post highlighted native MTP hitting 60 t/s at 130,000 context. But DFlash 2 just completely shattered that ceiling. By using parallel block diffusion drafting (predicting whole blocks of tokens in a single pass using dynamic convolutions), DFlash achieves a massive 5.39 token acceptance rate. THE ALPHA TWEAK: `n-max 7` eats too much VRAM for draft states. But if you drop the draft limit to `--spec-draft-n-max 4`, you slash the VRAM overhead and actually increase the throughput. Here is the new 24GB VRAM Physics Matrix (DFlash 2 @ n-max 4): - 30k Context: 1,725 t/s prefill | 87.05 t/s decode | 22.2 GB VRAM - 80k Context: 1,789 t/s prefill | 84.20 t/s decode | 23.3 GB VRAM - 110k Context: 1,767 t/s prefill | 83.35 t/s decode | 23.96 GB VRAM (110k context at 83+ tokens a second sitting exactly on the 24GB hardware limit is absolute wizardry). How to compile the PR today: git clone cd llama.cpp git fetch origin pull/27342/head:pr-27342 git switch pr-27342 cmake -B build -DGGML_CUDA=ON && cmake --build build -j Llama.cpp flags for Dflash (110k Context Ceiling): ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -md Qwen3.8-27B-DFlash2-Q4_K_M.gguf --spec-type draft-dflash --spec-draft-n-max 4 -c 110000 -ngl 99 --port 8080 -ctv q4_0 -ctk q4_0 The fact that the open source community is shipping block diffusion drafters so quickly that run entirely locally on a gaming GPU is unbelievable. If you own a single RTX 3090 or 4090, it is officially time to upgrade to qwen 3.8 27b with dflash 2 and cancel your API subscriptions and let local silicon eat the cloud. This model beats GPT 5.6 Terra, GLM 5.2 DeepSeek V4 Pro, Muse Spark 1.2 and Claude Opus 4.8 on the artificial analysis agentic index (details in the replies) Hugging Face GGUF links (Base + DFlash2) and the full visual VRAM scaling and Dflash2 vs MTP graphs are also in the replies below. are you sticking to native MTP for the 130k context, or sacrificing 20k context to redline your decode speed? How many tokens/sec are you pushing on your current local rig?

Alok

103,895 views • 17 days ago

The "I don't have enough VRAM" excuse just died. I’m running Meta’s new 30B Muse Glimmer Q6_K_XL with a massive 130k context window on just 26GB VRAM FREE compute on Kaggle. Kaggle provides you free 2x Nvidia T4 GPUs. 30 hours usage each week! Yesterday, I showed you the violent throughput of Muse Glimmer on a single RTX 4090. Today, we are securing a Dual NVIDIA T4 GPU cluster with 32GB of total VRAM for exactly $0 and dropping the massive 24.5GB Q6_K_XL GGUF onto it. Here is the exact Kaggle workflow and benchmarking breakdown: # 1. The Storage Bypass & Setup I built a clean cell by cell script in the file. We dynamically fetch the CUDA accelerated llama.cpp binaries and use wget to stream the model directly into Kaggle's /kaggle/tmp scratch storage, which cleanly bypasses their 19.5GB output directory limit. # 2. The Multi GPU Performance With the -ngl 99 flag offloading all model layers across both T4 GPUs (32GB VRAM combined), we pushed a massive 131,072 token context window (-c 131072). The benchmark numbers: Prefill: 265.9 t/s Decode: 9.0 t/s VRAM Total: 26.5 GB # 3. The Architecture Insight The Q6_K_XL model itself is 24.5 GB. Because of Muse Glimmer's aggressive 16:1 GQA, the unquantized KV cache for a massive 130k context window only takes up 2 GB of memory. No heavily degraded Q4 KV quantization required. It just works. No compiling from source. No credit card. No OOM crashes. Zero excuses. If you’re running a single RTX 3090, 4090, or 5090, you need to experience this hyper efficient KV cache right now before the upcoming Qwen 3.8 27B drop completely steals your VRAM tomorrow. pick the Q4 or Q5 quants for 24 GB VRAM rigs. I'm dropping the Unsloth huggingface GGUF links and the free Kaggle notebook link in the replies. spin up your own instance, and show me your multi GPU benchmarks.

The "I don't have enough VRAM" excuse just died. I’m running Meta’s new 30B Muse Glimmer Q6_K_XL with a massive 130k context window on just 26GB VRAM FREE compute on Kaggle. Kaggle provides you free 2x Nvidia T4 GPUs. 30 hours usage each week! Yesterday, I showed you the violent throughput of Muse Glimmer on a single RTX 4090. Today, we are securing a Dual NVIDIA T4 GPU cluster with 32GB of total VRAM for exactly $0 and dropping the massive 24.5GB Q6_K_XL GGUF onto it. Here is the exact Kaggle workflow and benchmarking breakdown: # 1. The Storage Bypass & Setup I built a clean cell by cell script in the file. We dynamically fetch the CUDA accelerated llama.cpp binaries and use wget to stream the model directly into Kaggle's /kaggle/tmp scratch storage, which cleanly bypasses their 19.5GB output directory limit. # 2. The Multi GPU Performance With the -ngl 99 flag offloading all model layers across both T4 GPUs (32GB VRAM combined), we pushed a massive 131,072 token context window (-c 131072). The benchmark numbers: Prefill: 265.9 t/s Decode: 9.0 t/s VRAM Total: 26.5 GB # 3. The Architecture Insight The Q6_K_XL model itself is 24.5 GB. Because of Muse Glimmer's aggressive 16:1 GQA, the unquantized KV cache for a massive 130k context window only takes up 2 GB of memory. No heavily degraded Q4 KV quantization required. It just works. No compiling from source. No credit card. No OOM crashes. Zero excuses. If you’re running a single RTX 3090, 4090, or 5090, you need to experience this hyper efficient KV cache right now before the upcoming Qwen 3.8 27B drop completely steals your VRAM tomorrow. pick the Q4 or Q5 quants for 24 GB VRAM rigs. I'm dropping the Unsloth huggingface GGUF links and the free Kaggle notebook link in the replies. spin up your own instance, and show me your multi GPU benchmarks.

Alok

19,089 views • 22 days ago

hyunsuk and jihoon snuck into the dorms at lunch scared the staff would catch them eating ramen in secret but got even more scared when the door opened but nobody was there + it wont close on its own as it should 😭 #최현석 #지훈 (at 11:52 of the content) 🦔 (heard the sound of door code pressed, door opening and shoes taken off but the sound stopped there and nobody came in) 🦔 we thought someone arrived… tbh that time we snuck out without the staff knowing so we were alr scared 🦔 “oh we cant get caught though” so we both froze for awhile and tbh when we were trainees, they were very strict… 🦔 we were gonna eat ramen in secret that time so we hid it and figured out what to do and when we settled down, no one came in. 🦔 (said they’d hear it if they went somewhere) 🦔 from then on, bc jihoon and i get scared so easily, were like “you go!” “no, you go” and argued for about 10 mins cos we get easily scared. 🦔 i eventually went out to check but there was no one near the shoe rack and the funny thing is it was just jihoon and i, just the two of us so there should only be two pairs of shoes, though there were untiedied shoes too, but there was another pair of shoes that weren’t there when we came in. so we wondered if someone came in quietly but in the end it’s really just the two of us inside the dorms. 🦔 and what was scarier was that like in most apartment buildings, the front door force closes so strongly so if you dont hold it open, it would automatically close right? ours is the same. it wasnt even left on a veryain angle but it stayed open… 🦔 so we spent another 10 mins arguing again and i went outside thinking someone must be holding on to it but there really was no one there. so we tried closing and opening it at that angle again and let go of it but it would close definitely. it doesnt make sense right? 🦔 so thats when i first understood it when people say when you’re truly scared, you cant even scream, you just freeze. 🦔 after we snapped out of it, we wer able to check the intercom footage, but nobody got off the elevator, nobody came up the stairs either, it just showed the door opening by itself like this so we were absolutely terrified and i think we left right away.

hyunsuk and jihoon snuck into the dorms at lunch scared the staff would catch them eating ramen in secret but got even more scared when the door opened but nobody was there + it wont close on its own as it should 😭 #최현석 #지훈 (at 11:52 of the content) 🦔 (heard the sound of door code pressed, door opening and shoes taken off but the sound stopped there and nobody came in) 🦔 we thought someone arrived… tbh that time we snuck out without the staff knowing so we were alr scared 🦔 “oh we cant get caught though” so we both froze for awhile and tbh when we were trainees, they were very strict… 🦔 we were gonna eat ramen in secret that time so we hid it and figured out what to do and when we settled down, no one came in. 🦔 (said they’d hear it if they went somewhere) 🦔 from then on, bc jihoon and i get scared so easily, were like “you go!” “no, you go” and argued for about 10 mins cos we get easily scared. 🦔 i eventually went out to check but there was no one near the shoe rack and the funny thing is it was just jihoon and i, just the two of us so there should only be two pairs of shoes, though there were untiedied shoes too, but there was another pair of shoes that weren’t there when we came in. so we wondered if someone came in quietly but in the end it’s really just the two of us inside the dorms. 🦔 and what was scarier was that like in most apartment buildings, the front door force closes so strongly so if you dont hold it open, it would automatically close right? ours is the same. it wasnt even left on a veryain angle but it stayed open… 🦔 so we spent another 10 mins arguing again and i went outside thinking someone must be holding on to it but there really was no one there. so we tried closing and opening it at that angle again and let go of it but it would close definitely. it doesnt make sense right? 🦔 so thats when i first understood it when people say when you’re truly scared, you cant even scream, you just freeze. 🦔 after we snapped out of it, we wer able to check the intercom footage, but nobody got off the elevator, nobody came up the stairs either, it just showed the door opening by itself like this so we were absolutely terrified and i think we left right away.

♡

41,317 views • 2 months ago

All this before they find the shooter. They just assume it couldn’t have been one of their own to further their agenda. I hated Charlie Kirk, I genuinely wouldn’t have pissed on him if he was on fire, but any idiot knows that killing him would only spark these idiots into thinking they were right, to push them further into their derangement and propaganda that they deep throat on the daily. Again, I hated Charlie Kirk, but for anyone on the left to kill him would be stupid. The American far right had so much more to gain by putting his lights out in such a public way. Charlie Kirk was wiped out by the very scum that be bathed in.

All this before they find the shooter. They just assume it couldn’t have been one of their own to further their agenda. I hated Charlie Kirk, I genuinely wouldn’t have pissed on him if he was on fire, but any idiot knows that killing him would only spark these idiots into thinking they were right, to push them further into their derangement and propaganda that they deep throat on the daily. Again, I hated Charlie Kirk, but for anyone on the left to kill him would be stupid. The American far right had so much more to gain by putting his lights out in such a public way. Charlie Kirk was wiped out by the very scum that be bathed in.

Bricktop_NAFO

1,638,655 views • 11 months ago

🚨 Shocking! If you think you've seen it all, just wait until you see this. Lando Norris's car is armed with extremely sharp objects on the front wing designed to puncture other cars' tires. Max Verstappen already had a puncture earlier, but watch how Lando (deliberately) steers into Max's rear tire to ensure he gets a puncture. With four teams now so close together, next time it could be Lewis Hamilton or Charles Leclerc. How can the FIA allow McLaren to race like this? And then you hear Stella and Zak Brown talking about 'fair racing' 🤨 FIA Oracle Red Bull Racing Max Verstappen Mercedes-AMG PETRONAS F1 Team Scuderia Ferrari HP George Russell Charles Leclerc Carlos Sainz Formula 1 Sergio Pérez Mohammed Ben Sulayem

🚨 Shocking! If you think you've seen it all, just wait until you see this. Lando Norris's car is armed with extremely sharp objects on the front wing designed to puncture other cars' tires. Max Verstappen already had a puncture earlier, but watch how Lando (deliberately) steers into Max's rear tire to ensure he gets a puncture. With four teams now so close together, next time it could be Lewis Hamilton or Charles Leclerc. How can the FIA allow McLaren to race like this? And then you hear Stella and Zak Brown talking about 'fair racing' 🤨 FIA Oracle Red Bull Racing Max Verstappen Mercedes-AMG PETRONAS F1 Team Scuderia Ferrari HP George Russell Charles Leclerc Carlos Sainz Formula 1 Sergio Pérez Mohammed Ben Sulayem

MV33Racing🏎

480,646 views • 2 years ago

I’m joining OpenAI Codex to work on the future of agentic development! At Cursor, I got to see the shift from autocomplete to agents. The next step isn’t a better IDE. It’s an Agent Development Environment (ADE): systems and tools for orchestrating agents, reasoning over their outputs, and making them autonomous enough to reliably complete ambitious work. After chatting with Alexander Embiricos and Tibo, it was clear that Codex is the best place to realize this vision. The team has consistently shipped SOTA models for agentic coding (check out gpt-5.3-codex) and I’m pumped for the future that the new Codex App points to. What I’m most excited about is the broader mission: accelerating the knowledge work economy. All agents are coding agents, and we’re already seeing Codex used across every job function within organizations. I’m extremely grateful for my time at Cursor, working with the incredible team, and I’m proud of what we built together. I’m excited to take an even bigger swing with Codex. If you’re curious to get a glimpse of where we are headed, download the Codex App! If you want to work on this mission, please apply or reach out - we are hiring across all functions! You can just build things.

I’m joining OpenAI Codex to work on the future of agentic development! At Cursor, I got to see the shift from autocomplete to agents. The next step isn’t a better IDE. It’s an Agent Development Environment (ADE): systems and tools for orchestrating agents, reasoning over their outputs, and making them autonomous enough to reliably complete ambitious work. After chatting with Alexander Embiricos and Tibo, it was clear that Codex is the best place to realize this vision. The team has consistently shipped SOTA models for agentic coding (check out gpt-5.3-codex) and I’m pumped for the future that the new Codex App points to. What I’m most excited about is the broader mission: accelerating the knowledge work economy. All agents are coding agents, and we’re already seeing Codex used across every job function within organizations. I’m extremely grateful for my time at Cursor, working with the incredible team, and I’m proud of what we built together. I’m excited to take an even bigger swing with Codex. If you’re curious to get a glimpse of where we are headed, download the Codex App! If you want to work on this mission, please apply or reach out - we are hiring across all functions! You can just build things.

Rohan Varma

759,796 views • 6 months ago