Загрузка видео...

Не удалось загрузить видео

На главную

oMLX is working really well as single machine inference engine for coding agents! Caching is managed perfectly (it can use a ton of disk space, be aware!) and oQ quantization delivers great results. Behind the scenes it uses the standard MLX building blocks (75% created by Prince Canuma 🙏):...

ivanfioravanti's profile picture

Ivan Fioravanti ᯅ

40,422 subscribers

19,420 просмотров • 6 месяцев назад •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

oMLX last version seems having some issues in decoding. 🧐 Here a test with Qwen3.6-35B-A3B-MLX-6bit on M5 Max using. 🥇 LM Studio MLX 1.8.5 → 100.9 toks/s 🥈 mlx-vlm 0.6.2 → 100.1 toks/s 🥉 oMLX 0.4.2 dev3 → 58.7 toks/s 👀 Avg Gen TPS: oMLX 58.7 → LM Studio 100.9 (+71.8%) I have to thank pymike00 that raised this oMLX issue after seeing my video on using it with Codex. I bet there is a bug in oMLX chat and server at the moment, because internal benchmarks are ok, video attached. I bet Jun Kim will fix this soon 💪
0:27

oMLX last version seems having some issues in decoding. 🧐 Here a test with Qwen3.6-35B-A3B-MLX-6bit on M5 Max using. 🥇 LM Studio MLX 1.8.5 → 100.9 toks/s 🥈 mlx-vlm 0.6.2 → 100.1 toks/s 🥉 oMLX 0.4.2 dev3 → 58.7 toks/s 👀 Avg Gen TPS: oMLX 58.7 → LM Studio 100.9 (+71.8%) I have to thank pymike00 that raised this oMLX issue after seeing my video on using it with Codex. I bet there is a bug in oMLX chat and server at the moment, because internal benchmarks are ok, video attached. I bet Jun Kim will fix this soon 💪

Ivan Fioravanti ᯅ

20,066 просмотров • 4 месяцев назад

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!
0:52

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!

Prince Canuma

82,349 просмотров • 6 месяцев назад

llamacpp vs omlx with Gemma-4-26B-A4B-it 🔥 🥇 llamacpp 🥈 oMLX Notes: - 11 parallel tasks - qat-Q4_0 vs OptiQ-4bit on omlx. - M5 Max 128GB macOS 27 Beta 1, relevant read below. - Fork of Gemma cookbook repo used with oMLX and iTerm2 usage added. Here: Observations: - llamacpp is more fluid in general, oMLX tends to slow down during prefill phase of a new request - I tested some MLX Gemma 4 qat, but they are all large in size, slower and with lower quality 🤷🏻‍♂️ - OptiQ quantization looks stronger than plain 4bit or 5bit MLX - Final result I slightly prefer OptiQ-4bit SVG gallery (on the right) - Results keep changing everytime I start a new instance of llamacpp or omlx, with differences up to 50%. Maybe due to macOS 27 beta1?
0:31

llamacpp vs omlx with Gemma-4-26B-A4B-it 🔥 🥇 llamacpp 🥈 oMLX Notes: - 11 parallel tasks - qat-Q4_0 vs OptiQ-4bit on omlx. - M5 Max 128GB macOS 27 Beta 1, relevant read below. - Fork of Gemma cookbook repo used with oMLX and iTerm2 usage added. Here: Observations: - llamacpp is more fluid in general, oMLX tends to slow down during prefill phase of a new request - I tested some MLX Gemma 4 qat, but they are all large in size, slower and with lower quality 🤷🏻‍♂️ - OptiQ quantization looks stronger than plain 4bit or 5bit MLX - Final result I slightly prefer OptiQ-4bit SVG gallery (on the right) - Results keep changing everytime I start a new instance of llamacpp or omlx, with differences up to 50%. Maybe due to macOS 27 beta1?

Ivan Fioravanti ᯅ

16,576 просмотров • 4 месяцев назад

Very quick comparison between Ornith 1.5 35B MoE and Qwen 3.8 27B Dense. 📣 Clearly it's not a fair one, but let's in any case see how it went. 397B download in progress! In the Videos below: - Brick SA -> Qwen 3.8 - Lego Streets -> Ornith 1.5 Context: - Pi agent used in both cases, prompt below - M3 Ultra 512GB - Ornith-1.5-35B-A3B-oQ8e MTP hosted on oMLX - mlx-community/Qwen3.8-27B-8bit DFlash 2 hosted on mlx-dspark - Ornith has been incredibly fast with speed from 75 t/s to 45 t/s (160K+ context) - Qwen3.8 suffered context much more reaching 5 t/s above 160K context, but it can be due to engine tested still work in progress - Prompt: using threejs, and cdn, create a lego like game that's inspired by gta san andreas, with beautiful aesthetics and graphics and ability to steal cars. it should be 3d and have a nice and large map and areas. the graphics should be decent and nice. it should feature iconic things from gta san andreas Notes: - Qwen 3.8 result is 0-shot, while Ornith 1.5 required 6 iterations - Ornith 1.5 is not at the same level of autonomy as Qwen 3.8 27B honestly. To try getting the same results I'm constantly nudging, steering and 🤬 at it. - Qwen 3.8 took 6 hours to complete, but it was more a problem of timeout of Chrome headless used for testing, real prefill/decode TBD. I'll test again now with oMLX - Pi agent has a nearly perfect Cache Hit ratio that for local models is MEGA important!
0:31

Very quick comparison between Ornith 1.5 35B MoE and Qwen 3.8 27B Dense. 📣 Clearly it's not a fair one, but let's in any case see how it went. 397B download in progress! In the Videos below: - Brick SA -> Qwen 3.8 - Lego Streets -> Ornith 1.5 Context: - Pi agent used in both cases, prompt below - M3 Ultra 512GB - Ornith-1.5-35B-A3B-oQ8e MTP hosted on oMLX - mlx-community/Qwen3.8-27B-8bit DFlash 2 hosted on mlx-dspark - Ornith has been incredibly fast with speed from 75 t/s to 45 t/s (160K+ context) - Qwen3.8 suffered context much more reaching 5 t/s above 160K context, but it can be due to engine tested still work in progress - Prompt: using threejs, and cdn, create a lego like game that's inspired by gta san andreas, with beautiful aesthetics and graphics and ability to steal cars. it should be 3d and have a nice and large map and areas. the graphics should be decent and nice. it should feature iconic things from gta san andreas Notes: - Qwen 3.8 result is 0-shot, while Ornith 1.5 required 6 iterations - Ornith 1.5 is not at the same level of autonomy as Qwen 3.8 27B honestly. To try getting the same results I'm constantly nudging, steering and 🤬 at it. - Qwen 3.8 took 6 hours to complete, but it was more a problem of timeout of Chrome headless used for testing, real prefill/decode TBD. I'll test again now with oMLX - Pi agent has a nearly perfect Cache Hit ratio that for local models is MEGA important!

Ivan Fioravanti ᯅ

12,150 просмотров • 2 месяцев назад

SAM-2 is apache license - you can indeed just build crazy apps with them without fear. This weekend I ported SAM-2 models to run optimally on Apple Silicon. I have also quantized all the SAM models to 16bit, 8bit, and 4bit variants (collection below) Repo is open source and pip installable. Comes with all you'll need for interactive video segmentation - - bidirectional propagation, - positive and negative clicks, - bounding boxes, - streaming masks back to user, - lightweight fast-api server that you can extend I basically asked Codex to port over the original Facebook sam2 codebase to mlx and optimize it for mlx. And it just worked. This demo for example is a 4bit base-plus model (~100MB download) To speed up, you can also do spatial or temporal downsampling. That speeds inference on my old macbook by ~3x and the hit in parity is negligible.
0:53

SAM-2 is apache license - you can indeed just build crazy apps with them without fear. This weekend I ported SAM-2 models to run optimally on Apple Silicon. I have also quantized all the SAM models to 16bit, 8bit, and 4bit variants (collection below) Repo is open source and pip installable. Comes with all you'll need for interactive video segmentation - - bidirectional propagation, - positive and negative clicks, - bounding boxes, - streaming masks back to user, - lightweight fast-api server that you can extend I basically asked Codex to port over the original Facebook sam2 codebase to mlx and optimize it for mlx. And it just worked. This demo for example is a 4bit base-plus model (~100MB download) To speed up, you can also do spatial or temporal downsampling. That speeds inference on my old macbook by ~3x and the hit in parity is negligible.

AVB

17,428 просмотров • 5 месяцев назад

I used Gemma4 + Falcon Perception from this mlx-vlm release to build a grounded reasoning agent runs fully local on M3 the idea: VLMs are great at reasoning but not great at measuring. Falcon Perception is great at segmentation but cant reason. so you loop them: Gemma4 decides what to look for, FP segments it and returns pixel-accurate coordinates, Gemma4 reasons on the numbers ask "is the blue player offside?" → it grounds the players, finds the second-to-last defender, compares centroid positions, applies the rule. check the video for some examples Prince Canuma I can submit a PR with this demo if you want
0:38

I used Gemma4 + Falcon Perception from this mlx-vlm release to build a grounded reasoning agent runs fully local on M3 the idea: VLMs are great at reasoning but not great at measuring. Falcon Perception is great at segmentation but cant reason. so you loop them: Gemma4 decides what to look for, FP segments it and returns pixel-accurate coordinates, Gemma4 reasons on the numbers ask "is the blue player offside?" → it grounds the players, finds the second-to-last defender, compares centroid positions, applies the rule. check the video for some examples Prince Canuma I can submit a PR with this demo if you want

Yasser Dahou

110,318 просмотров • 6 месяцев назад

just built a local AI waifu on my two macbooks by using the oss parallax inference library!!! not super smooth yet but parallax is still the easiest way to run llms locally on top of mlx - and no bigco can steal my waifu. still need folks to help her see/hear/draw since mlx has no legit wrapper yet. i’m running a campaign giving away mac minis + dgx spark. please please share your localhost ai model setup by using parallax (you don't need a gpu to try self-hosting you can just start with an extra macbook you borrowed from your friend like i did) or any fun app built on top of yourself host model to link below. come build sovereign intelligence with me 🫶
7:41

just built a local AI waifu on my two macbooks by using the oss parallax inference library!!! not super smooth yet but parallax is still the easiest way to run llms locally on top of mlx - and no bigco can steal my waifu. still need folks to help her see/hear/draw since mlx has no legit wrapper yet. i’m running a campaign giving away mac minis + dgx spark. please please share your localhost ai model setup by using parallax (you don't need a gpu to try self-hosting you can just start with an extra macbook you borrowed from your friend like i did) or any fun app built on top of yourself host model to link below. come build sovereign intelligence with me 🫶

Asuka Zheng🎀@ICML

178,998 просмотров • 10 месяцев назад

🤯This is crazy... Qwen 3.8 Flash Next IS really good... it basically one shot this, then I just told it to add the MLX-Serve / Qwen structure after. (mixed 4-8bit version) (~90gb ram, it did use full 256k ctx) Prompt: build me a minecraft clone using threejs, npm, typescript, test files, linting, and better graphics than minecraft, make it look good. Just simple block building with the basic minecraft blocks, but make it look really good with amazing graphics. Generated textures and sounds. Benchmaxxing ? Maybe, needs more testing, but I have been seeing some good results on Youtube also.
1:47

🤯This is crazy... Qwen 3.8 Flash Next IS really good... it basically one shot this, then I just told it to add the MLX-Serve / Qwen structure after. (mixed 4-8bit version) (~90gb ram, it did use full 256k ctx) Prompt: build me a minecraft clone using threejs, npm, typescript, test files, linting, and better graphics than minecraft, make it look good. Just simple block building with the basic minecraft blocks, but make it look really good with amazing graphics. Generated textures and sounds. Benchmaxxing ? Maybe, needs more testing, but I have been seeing some good results on Youtube also.

David Dalcu

26,075 просмотров • 1 месяц назад

This inference engine runs LLMs 4x faster than llama.cpp: (Qwen3.5-9B at 92 tokens/s on 16GB Mac) I ran Qwen3.5 9B on a base M5 MacBook Pro using three local inference engines, one at a time. → llama.cpp reached 22.0 tokens/s → MLX reached 25.1 tokens/s → Uzu reached 92.1 tokens/s The video shows all three runs side by side. The first two results do not mean llama.cpp or MLX are poorly optimized. Both are approaching a physical limit of ordinary local decoding. Generating each new token normally requires another pass through the model. That pass streams roughly 5.2 GB of weights through memory. A base M5 provides about 153 GB/s of memory bandwidth. At one token per pass, the theoretical ceiling is roughly 153 ÷ 5.2, or 29 tokens/s, however fast the GPU computes. Uzu gets past that ceiling by optimizing speculative decoding for Apple Silicon. Speculative decoding uses a smaller drafting system to propose several future tokens. The full 9B model then checks those proposals together instead of generating every token through a separate pass. Uzu pushes this further in two ways. → It drafts up to 16 positions and uses Weaver to turn independent guesses into coherent sequences. → It verifies several possible sequences as a tree, then commits state only for the path the full model accepts. The 9B model remains in control of the final output. Uzu simply gives it more useful work to approve in each pass. In my run, every pass of the 9B model produced 7.5 output tokens on average. Ordinary decoding produces exactly 1. That is how Uzu reached 92.1 tokens/s on hardware whose one-token-per-pass ceiling is roughly 29. The rest of the engine is designed around this path too. Uzu ships its own quantized checkpoints, and on M5 chips its Metal kernels run verification through the GPU’s accelerated int8 matrix path. It is an open-source inference engine from Mirai, with bindings for Rust, Swift, Python, and TypeScript. The repository is here: (don't forget to star 🌟) I wrote the full breakdown of why local inference is memory-bound and how Uzu gets past that limit, with reproducible results and code. The article is quoted below.
0:26

This inference engine runs LLMs 4x faster than llama.cpp: (Qwen3.5-9B at 92 tokens/s on 16GB Mac) I ran Qwen3.5 9B on a base M5 MacBook Pro using three local inference engines, one at a time. → llama.cpp reached 22.0 tokens/s → MLX reached 25.1 tokens/s → Uzu reached 92.1 tokens/s The video shows all three runs side by side. The first two results do not mean llama.cpp or MLX are poorly optimized. Both are approaching a physical limit of ordinary local decoding. Generating each new token normally requires another pass through the model. That pass streams roughly 5.2 GB of weights through memory. A base M5 provides about 153 GB/s of memory bandwidth. At one token per pass, the theoretical ceiling is roughly 153 ÷ 5.2, or 29 tokens/s, however fast the GPU computes. Uzu gets past that ceiling by optimizing speculative decoding for Apple Silicon. Speculative decoding uses a smaller drafting system to propose several future tokens. The full 9B model then checks those proposals together instead of generating every token through a separate pass. Uzu pushes this further in two ways. → It drafts up to 16 positions and uses Weaver to turn independent guesses into coherent sequences. → It verifies several possible sequences as a tree, then commits state only for the path the full model accepts. The 9B model remains in control of the final output. Uzu simply gives it more useful work to approve in each pass. In my run, every pass of the 9B model produced 7.5 output tokens on average. Ordinary decoding produces exactly 1. That is how Uzu reached 92.1 tokens/s on hardware whose one-token-per-pass ceiling is roughly 29. The rest of the engine is designed around this path too. Uzu ships its own quantized checkpoints, and on M5 chips its Metal kernels run verification through the GPU’s accelerated int8 matrix path. It is an open-source inference engine from Mirai, with bindings for Rust, Swift, Python, and TypeScript. The repository is here: (don't forget to star 🌟) I wrote the full breakdown of why local inference is memory-bound and how Uzu gets past that limit, with reproducible results and code. The article is quoted below.

Akshay 🚀

81,359 просмотров • 11 дней назад

Qwen3.8-Flash-Next is still going strong at 364.7K tokens of context on an M5 Max. And this isn’t just a static long-context test. The model was reasoning about how to speed up its own workflow while using tools, and the tool calls kept working without misses. Setup: • Qwen3.8-Flash-Next • M5 Max • 128GB unified memory • MLX-Serve PR #363 • OpenCode 2 • 364.7K context The interesting part isn’t simply getting hundreds of thousands of tokens into memory. It’s what happens once the context gets this large. Long-context inference usually comes with a painful tradeoff. As the KV cache grows, memory pressure increases and generation can slow down. But this setup is still pushing through 364K tokens while maintaining a usable agent workflow. The model can reason, call tools, inspect results, continue working, and keep the session moving. And the tool calls reportedly haven’t missed so far. That’s important for agentic coding. A huge context window is only useful if the model can actually operate reliably inside it. A 400K-token context that constantly breaks tool calls isn’t very useful. A 364K session that can keep reasoning and executing tools is a different story. And the test isn’t finished yet. The current run is approaching 400K tokens, with the expectation that it can keep going. This is also another interesting example of why Apple Silicon keeps showing up in local LLM experiments. The M5 Max’s unified memory gives a large model and its growing KV cache access to one shared memory pool. With MLX-Serve continuing to improve, these machines are becoming surprisingly capable long-context inference boxes. The bigger takeaway: Context length is becoming a workload, not just a model specification. Running a model at 256K is one thing. Keeping an agent alive at 300K+ while it reasons and uses tools is much more interesting. And Qwen3.8-Flash-Next is showing that this can be pushed surprisingly far on a single 128GB Mac. 364.7K and counting. Next stop: 400K.
0:33

Qwen3.8-Flash-Next is still going strong at 364.7K tokens of context on an M5 Max. And this isn’t just a static long-context test. The model was reasoning about how to speed up its own workflow while using tools, and the tool calls kept working without misses. Setup: • Qwen3.8-Flash-Next • M5 Max • 128GB unified memory • MLX-Serve PR #363 • OpenCode 2 • 364.7K context The interesting part isn’t simply getting hundreds of thousands of tokens into memory. It’s what happens once the context gets this large. Long-context inference usually comes with a painful tradeoff. As the KV cache grows, memory pressure increases and generation can slow down. But this setup is still pushing through 364K tokens while maintaining a usable agent workflow. The model can reason, call tools, inspect results, continue working, and keep the session moving. And the tool calls reportedly haven’t missed so far. That’s important for agentic coding. A huge context window is only useful if the model can actually operate reliably inside it. A 400K-token context that constantly breaks tool calls isn’t very useful. A 364K session that can keep reasoning and executing tools is a different story. And the test isn’t finished yet. The current run is approaching 400K tokens, with the expectation that it can keep going. This is also another interesting example of why Apple Silicon keeps showing up in local LLM experiments. The M5 Max’s unified memory gives a large model and its growing KV cache access to one shared memory pool. With MLX-Serve continuing to improve, these machines are becoming surprisingly capable long-context inference boxes. The bigger takeaway: Context length is becoming a workload, not just a model specification. Running a model at 256K is one thing. Keeping an agent alive at 300K+ while it reasons and uses tools is much more interesting. And Qwen3.8-Flash-Next is showing that this can be pushed surprisingly far on a single 128GB Mac. 364.7K and counting. Next stop: 400K.

FHILY👑

39,982 просмотров • 1 месяц назад

A voice agent powered by gpt-oss. Running locally on my macBook. Demo recorded in a Waymo with WiFi turned off. I'm still on my space game voice AI kick, obviously. Code link below. For conversational voice AI, you want to set the gpt-oss reasoning behavior to "low". (The default is "medium".) Notes on how to do that and a jinja template you can use are in the repo. The LLM in the demo video is the big, 120B version of gpt-oss. You can use the smaller, 20B model for this, of course. But OpenAI really did a cool thing here designing the 120B model to run in "just" 80GB of VRAM. And the llama.cpp mlx inference is fast: ~250ms TTFT. Running a big model on-device feels like a time warp into the future of AI.
4:35

A voice agent powered by gpt-oss. Running locally on my macBook. Demo recorded in a Waymo with WiFi turned off. I'm still on my space game voice AI kick, obviously. Code link below. For conversational voice AI, you want to set the gpt-oss reasoning behavior to "low". (The default is "medium".) Notes on how to do that and a jinja template you can use are in the repo. The LLM in the demo video is the big, 120B version of gpt-oss. You can use the smaller, 20B model for this, of course. But OpenAI really did a cool thing here designing the 120B model to run in "just" 80GB of VRAM. And the llama.cpp mlx inference is fast: ~250ms TTFT. Running a big model on-device feels like a time warp into the future of AI.

kwindla

202,309 просмотров • 1 год назад

I made a few updates to the MLX port of F5 TTS over the holiday: - Longform generation is now supported, and text will be split automatically on sentence boundaries for natural-sounding speech. - You can now pipe in text from another process (e.g. a language model) and listen to the speech as each segment is generated. - I implemented RK4 sampling for the ODE, which achieves much better quality with fewer steps — it's now possible to generate speech at 1.6x realtime on my M3 Max. See the video of an example of generating speech in realtime from a language model entirely on-device! 🚀
0:54

I made a few updates to the MLX port of F5 TTS over the holiday: - Longform generation is now supported, and text will be split automatically on sentence boundaries for natural-sounding speech. - You can now pipe in text from another process (e.g. a language model) and listen to the speech as each segment is generated. - I implemented RK4 sampling for the ODE, which achieves much better quality with fewer steps — it's now possible to generate speech at 1.6x realtime on my M3 Max. See the video of an example of generating speech in realtime from a language model entirely on-device! 🚀

Lucas Newman

22,998 просмотров • 1 год назад

It takes 5 minutes to create a world-class web research agent that runs 24/7 and returns structured search results. 1. You can limit the sources it's allowed to use 2. You can specify the goals it needs to complete 3. You can design the format structure of the results 4. You can configure how much computing power it will use I show you how to set up a Web Search Agent in the video. A few benefits here: 1. This is great for complex research tasks 2. It's really easy to add this like a LEGO piece to a more complex workflow 3. Create an agent once, and reuse it with different inputs 4. The agent adapts to your use case as you work with it By the way, Nimble published a whole library of templates and research agents you can reuse and modify. You can start sending queries to one of these existing agents, and it will begin to adapt to Go here to try it out: Thanks to the Nimble team for partnering with me on this post.
7:37

It takes 5 minutes to create a world-class web research agent that runs 24/7 and returns structured search results. 1. You can limit the sources it's allowed to use 2. You can specify the goals it needs to complete 3. You can design the format structure of the results 4. You can configure how much computing power it will use I show you how to set up a Web Search Agent in the video. A few benefits here: 1. This is great for complex research tasks 2. It's really easy to add this like a LEGO piece to a more complex workflow 3. Create an agent once, and reuse it with different inputs 4. The agent adapts to your use case as you work with it By the way, Nimble published a whole library of templates and research agents you can reuse and modify. You can start sending queries to one of these existing agents, and it will begin to adapt to Go here to try it out: Thanks to the Nimble team for partnering with me on this post.

Santiago

21,079 просмотров • 2 месяцев назад

“To be truly great you must be fearless and try things that may not work...” Cooked up this painting of the Fontaine de la Place du Général de Gaulle in Canne-France🇫🇷 and filmed it with my RayBanMeta glasses for the first time. It was really easy to use and quite convenient as well. And as you can see for yourself the video quality is great. Love it! Tell me what you think about the painting. 🎥: @the.eventguy Ambassade de France au Ghana LIONS | The Home of Creativity Meta @woofstudiosafrica
0:58

“To be truly great you must be fearless and try things that may not work...” Cooked up this painting of the Fontaine de la Place du Général de Gaulle in Canne-France🇫🇷 and filmed it with my RayBanMeta glasses for the first time. It was really easy to use and quite convenient as well. And as you can see for yourself the video quality is great. Love it! Tell me what you think about the painting. 🎥: @the.eventguy Ambassade de France au Ghana LIONS | The Home of Creativity Meta @woofstudiosafrica

ENIL

25,452 просмотров • 1 год назад

AI content creation should make things easier, not more complicated, and MiniMax Design really gets that It saves me so much time and effort, and somehow it’s still finding ways to surprise me I already thought it worked really well with MiniMax H3 But then MiniMax H3 Max Turbo showed up… And I was like, wait, what?! The speed is actually kind of crazy A 15s video can be generated in around 20s And then I saw the price Starting at just $0.018/s Like… seriously?! But what makes this even more useful is that MiniMax Design also has GPT Image 2.5 I can generate the visuals with GPT Image 2.5, keep the quality and consistency I want, then take those images straight into H3 Max Turbo and turn them into video Image → H3 Max Turbo → Video Fast images, fast video, and I don’t have to keep rebuilding the idea across different tools And apparently MiniMax Design wasn’t done — GPT-6 Astra is also available now at 92% off Honestly, MiniMax Design + GPT Image 2.5 + H3 Max Turbo is definitely going on my list of workflows I’ll be using a lot Every time I think MiniMax Design is done surprising me, there’s something else to try
0:29

AI content creation should make things easier, not more complicated, and MiniMax Design really gets that It saves me so much time and effort, and somehow it’s still finding ways to surprise me I already thought it worked really well with MiniMax H3 But then MiniMax H3 Max Turbo showed up… And I was like, wait, what?! The speed is actually kind of crazy A 15s video can be generated in around 20s And then I saw the price Starting at just $0.018/s Like… seriously?! But what makes this even more useful is that MiniMax Design also has GPT Image 2.5 I can generate the visuals with GPT Image 2.5, keep the quality and consistency I want, then take those images straight into H3 Max Turbo and turn them into video Image → H3 Max Turbo → Video Fast images, fast video, and I don’t have to keep rebuilding the idea across different tools And apparently MiniMax Design wasn’t done — GPT-6 Astra is also available now at 92% off Honestly, MiniMax Design + GPT Image 2.5 + H3 Max Turbo is definitely going on my list of workflows I’ll be using a lot Every time I think MiniMax Design is done surprising me, there’s something else to try

Markandey Sharma

41,856 просмотров • 1 месяц назад

Sam Altman reveals he deliberately got himself addicted to TikTok while building Sora, then deleted it because it was too powerful like a drug. "I deleted TikTok because it was just too powerful. Here's a crazy thing that happened. When we were building the Sora app, I made myself get addicted to TikTok because I wanted to learn." "I'd never really used it before... And I loved it. I love TikTok, I really thought it was great." "And then I thought I could control it. I was like, oh, it's actually kind of fun, but I only use it for 10 minutes to wind down before bed and I'm totally in control of it." "And then it was an hour one night, and then some Saturday afternoon I was on the couch for like 3 hours, and I was like, this is really not what I thought the iPhone was supposed to be about." "Now I'm really enjoying it in the moment like a drug, but I can tell it's bad for me. And then I briefly got it back under control, down to like 5 or 10 minutes a night. Then I was just like, enough."
1:04

Sam Altman reveals he deliberately got himself addicted to TikTok while building Sora, then deleted it because it was too powerful like a drug. "I deleted TikTok because it was just too powerful. Here's a crazy thing that happened. When we were building the Sora app, I made myself get addicted to TikTok because I wanted to learn." "I'd never really used it before... And I loved it. I love TikTok, I really thought it was great." "And then I thought I could control it. I was like, oh, it's actually kind of fun, but I only use it for 10 minutes to wind down before bed and I'm totally in control of it." "And then it was an hour one night, and then some Saturday afternoon I was on the couch for like 3 hours, and I was like, this is really not what I thought the iPhone was supposed to be about." "Now I'm really enjoying it in the moment like a drug, but I can tell it's bad for me. And then I briefly got it back under control, down to like 5 or 10 minutes a night. Then I was just like, enough."

Fireside Alpha

605,308 просмотров • 2 месяцев назад

fun weekend project: AutoJev. i was curious to see if i could train a competitive Jev-like model completely autonomously with a swarm of agents using our internal system. turns out you can get quite far! some details: - gave the swarm a devbox with an h200 gpu - the swarm is a mix of astra and fable; used sol and luna for synthetic data and filtering; ran for 20 hours - spent $3.1k in total ($1.9k on agents, $1.2k on data) - used qwen3.8-27b as the backbone. it underperforms Jev initially, but after training it is competitive/better, and it also supports 260k context and multimodal input - the agents tried RL, but it hasn't worked out so far, so they fell back to SFT on high-quality synthetic data - both accuracy and calibration are better - jev-compatible api, playground, and inference. on an h200 it gets around 120ms p95. that will be higher under contention, but still quite fast for an unoptimized version code: model: going to let it run longer, and will also bring up inference so folks can try it out.
0:40

fun weekend project: AutoJev. i was curious to see if i could train a competitive Jev-like model completely autonomously with a swarm of agents using our internal system. turns out you can get quite far! some details: - gave the swarm a devbox with an h200 gpu - the swarm is a mix of astra and fable; used sol and luna for synthetic data and filtering; ran for 20 hours - spent $3.1k in total ($1.9k on agents, $1.2k on data) - used qwen3.8-27b as the backbone. it underperforms Jev initially, but after training it is competitive/better, and it also supports 260k context and multimodal input - the agents tried RL, but it hasn't worked out so far, so they fell back to SFT on high-quality synthetic data - both accuracy and calibration are better - jev-compatible api, playground, and inference. on an h200 it gets around 120ms p95. that will be higher under contention, but still quite fast for an unoptimized version code: model: going to let it run longer, and will also bring up inference so folks can try it out.

Denis Yarats

28,497 просмотров • 28 дней назад