Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

mlx_lm server worked flawlessly with Qwen3.6-35B-A3B-8bit, and on M5 Max, the much faster prefill, gives a very pleasant coding experience. Here two OpenCode instances working on mlx_lm and mlx_vlm source code. Video in normal speed.

ivanfioravanti's profile picture

Ivan Fioravanti ᯅ

39,484 subscribers

19,671 Aufrufe • vor 3 Monaten •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

Video on LM Studio MLX v.1.8.5 + Codex + qwen3.6-35b-a3b 6bit on an M5 Max 128GB. Video is in normal speed to give you an idea of the experience. Command used: codex --oss -m qwen/qwen3.6-35b-a3b Local AI is becoming powerful 💪
1:18

Video on LM Studio MLX v.1.8.5 + Codex + qwen3.6-35b-a3b 6bit on an M5 Max 128GB. Video is in normal speed to give you an idea of the experience. Command used: codex --oss -m qwen/qwen3.6-35b-a3b Local AI is becoming powerful 💪

Ivan Fioravanti ᯅ

40,565 Aufrufe • vor 2 Monaten

Qwen3.6-35B-A3B-NVFP4 Benchmark: MoE vs. Dense Performance Report This report details a performance comparisonbetween two artificial intelligence models, focusing on the Qwen3.6-35B-A3B-NVFP4 and its predecessor. While both models achieved flawless accuracy across a comprehensive 65-test benchmark covering vision, video, and reasoning, the 35B version demonstrated significant speed advantages. Utilizing a Mixture of Experts architecture, the 35B model achieved nearly four times higher throughput and drastically reduced latency for initial responses. It also showed superior efficiency in speculative decoding and successfully managed massive context lengths up to 128,000 tokens. Ultimately, the source recommends the 35B model for high-speed production environments, while noting the 27B model remains more stable for specific complex reasoning modes.
1:57

Qwen3.6-35B-A3B-NVFP4 Benchmark: MoE vs. Dense Performance Report This report details a performance comparisonbetween two artificial intelligence models, focusing on the Qwen3.6-35B-A3B-NVFP4 and its predecessor. While both models achieved flawless accuracy across a comprehensive 65-test benchmark covering vision, video, and reasoning, the 35B version demonstrated significant speed advantages. Utilizing a Mixture of Experts architecture, the 35B model achieved nearly four times higher throughput and drastically reduced latency for initial responses. It also showed superior efficiency in speculative decoding and successfully managed massive context lengths up to 128,000 tokens. Ultimately, the source recommends the 35B model for high-speed production environments, while noting the 27B model remains more stable for specific complex reasoning modes.

Mike Gannotti

10,457 Aufrufe • vor 1 Monat

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 🙏): - mlx-lm - mlx-vlm - mlx-embeddings - mlx-audio I tested Qwen3.6-35B-A3B-oQ6 on M5 Max with two pi instances and it was fast and furious and leveraging cache like crazy as you can see in the video. Let me try to create some oQ versions (2,4,6?) of MiniMax M2.7 now and then I'll pass to distributed inference. I must win! 💪
3:57

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 🙏): - mlx-lm - mlx-vlm - mlx-embeddings - mlx-audio I tested Qwen3.6-35B-A3B-oQ6 on M5 Max with two pi instances and it was fast and furious and leveraging cache like crazy as you can see in the video. Let me try to create some oQ versions (2,4,6?) of MiniMax M2.7 now and then I'll pass to distributed inference. I must win! 💪

Ivan Fioravanti ᯅ

19,420 Aufrufe • vor 3 Monaten

Everyone's comparing the DGX Spark to a 5090 and calling it slow. I think that's the wrong comparison. I ran Qwen3.6 35B-A3B FP8 with the full 262K context window enabled (~96GB RAM) — something gaming GPUs can't really do. Results: 🟢No context: 51.3 tok/s, TTFT 110ms 🟣200K prefill: 34.6 tok/s, TTFT 85s (~2,341 tok/s prefill) Prefill is way faster than a Mac. And 35 tok/s deep into 200K context, on a model this strong, is genuinely usable. The Spark plays a different game.
1:04

Everyone's comparing the DGX Spark to a 5090 and calling it slow. I think that's the wrong comparison. I ran Qwen3.6 35B-A3B FP8 with the full 262K context window enabled (~96GB RAM) — something gaming GPUs can't really do. Results: 🟢No context: 51.3 tok/s, TTFT 110ms 🟣200K prefill: 34.6 tok/s, TTFT 85s (~2,341 tok/s prefill) Prefill is way faster than a Mac. And 35 tok/s deep into 200K context, on a model this strong, is genuinely usable. The Spark plays a different game.

stevibe

34,141 Aufrufe • vor 3 Monaten

First video about MTPLX it works! 🔥 Gonna do my context benchmarks and some coding on it ASAP! In my very first test here: 🥇 M3 Ultra on the right: 33 --> 68 tps 🥈 M5 Max on the left: 27 --> 55 tps But M5 Max has 2 external monitors connected and is doing ton of stuff in parallel. Amazing job by Youssof Al Toukhi here!
2:12

First video about MTPLX it works! 🔥 Gonna do my context benchmarks and some coding on it ASAP! In my very first test here: 🥇 M3 Ultra on the right: 33 --> 68 tps 🥈 M5 Max on the left: 27 --> 55 tps But M5 Max has 2 external monitors connected and is doing ton of stuff in parallel. Amazing job by Youssof Al Toukhi here!

Ivan Fioravanti ᯅ

14,267 Aufrufe • vor 3 Monaten

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 Aufrufe • vor 2 Monaten

You can use a 100% open source and MUCH cheaper/free alternative to Claude Code and Opus 4.5 OpenCode + MiniMax M2.1 can even build a 3D website using pure vibe coding. Steps are really simple: 1. Install OpenCode using the command 'npm i -g opencode-ai' 2. Get your MiniMax API key here: 3. Configure the MiniMax (official) mode Just type "opencode connect minimax" Coding plans start at $2… 10x cheaper than Claude Code (yes). You can also invite friends so they can get 10% off and you’ll get 10% API credits. (You can also use it locally if you have the config) And you're ready to build and iterate almost endlessly since the model is both way faster and cheaper than Opus in CC.
12:07

You can use a 100% open source and MUCH cheaper/free alternative to Claude Code and Opus 4.5 OpenCode + MiniMax M2.1 can even build a 3D website using pure vibe coding. Steps are really simple: 1. Install OpenCode using the command 'npm i -g opencode-ai' 2. Get your MiniMax API key here: 3. Configure the MiniMax (official) mode Just type "opencode connect minimax" Coding plans start at $2… 10x cheaper than Claude Code (yes). You can also invite friends so they can get 10% off and you’ll get 10% API credits. (You can also use it locally if you have the config) And you're ready to build and iterate almost endlessly since the model is both way faster and cheaper than Opus in CC.

Paul Couvert

42,443 Aufrufe • vor 7 Monaten

Let's be honest Claude Code was not working good with local LLMs. So, I Tried using OpenCode with LM Studio using Qwen 3.5 9B model on my RTX 5080 GPU. It's very usable and much faster than Claude Code. I'm actually planning to use this setup to build an entire vibe coded SaaS
2:31

Let's be honest Claude Code was not working good with local LLMs. So, I Tried using OpenCode with LM Studio using Qwen 3.5 9B model on my RTX 5080 GPU. It's very usable and much faster than Claude Code. I'm actually planning to use this setup to build an entire vibe coded SaaS

Tushar Verma

137,536 Aufrufe • vor 5 Monaten

here's what i vibecoded today: punchingface 🥊 an app to make Hugging Face models fight each other on coding and canvas challenges, built with qwen3.6 35b a3b in 24 hours! benchmarks numbers don't mean anything anymore, we need a way to visualize what the models are actually capable of, and canvas are one the best way to showcase it imo. why? because one single error in the code and everything breaks it shows the differences in a matter of seconds, way easier than manually reviewing the quality of the code produced for a complex project; much needed in the space with all the new finetunes dropping everyday! i did a little demo here with qwen3.6 vs qwopus glm 18b merged, the frankenstein model from Kyle Hessling the winner is clear here, qwen is crazy good and has nothing to prove. that said, qwopus 18b isn't terrible at all; the result isn’t the prettiest to the eye, but hey… it works! i've seen so many models just output a blank page (completely non-working code) so this is already a win frankenstein talks and thinks but he needs some extra brain surgery 🧠 results were expected (it's a very experimental model) but love the effort in the 18b direction from jackrong and kyle! the app was entirely vibecoded with qwen3.6, i didn't edit a single file manually. i can say with confidence that it really has the intelligence of claude sonnet 4.5 at a speed of 125tok/s on an rtx 5080 which models should i make fight next?
0:52

here's what i vibecoded today: punchingface 🥊 an app to make Hugging Face models fight each other on coding and canvas challenges, built with qwen3.6 35b a3b in 24 hours! benchmarks numbers don't mean anything anymore, we need a way to visualize what the models are actually capable of, and canvas are one the best way to showcase it imo. why? because one single error in the code and everything breaks it shows the differences in a matter of seconds, way easier than manually reviewing the quality of the code produced for a complex project; much needed in the space with all the new finetunes dropping everyday! i did a little demo here with qwen3.6 vs qwopus glm 18b merged, the frankenstein model from Kyle Hessling the winner is clear here, qwen is crazy good and has nothing to prove. that said, qwopus 18b isn't terrible at all; the result isn’t the prettiest to the eye, but hey… it works! i've seen so many models just output a blank page (completely non-working code) so this is already a win frankenstein talks and thinks but he needs some extra brain surgery 🧠 results were expected (it's a very experimental model) but love the effort in the 18b direction from jackrong and kyle! the app was entirely vibecoded with qwen3.6, i didn't edit a single file manually. i can say with confidence that it really has the intelligence of claude sonnet 4.5 at a speed of 125tok/s on an rtx 5080 which models should i make fight next?

left curve dev

19,153 Aufrufe • vor 3 Monaten

Introducing Kimi 2.6 Code. A Claude Code-like terminal experience built specifically for Kimi K2.6, effectively making it one of the most powerful open-source coding agents on the planet. Simply bring your API key and use /login. Repo here 👇
0:40

Introducing Kimi 2.6 Code. A Claude Code-like terminal experience built specifically for Kimi K2.6, effectively making it one of the most powerful open-source coding agents on the planet. Simply bring your API key and use /login. Repo here 👇

Pietro Schirano

135,945 Aufrufe • vor 3 Monaten

⚡️Live: You don't always need to use 3D elements in #threejs. In this example, a simple plane with a normal/displacement texture and lights that follows the mouse gives nice effects😍 The source code is here: #threejsJourney #r3f
1:40

⚡️Live: You don't always need to use 3D elements in #threejs. In this example, a simple plane with a normal/displacement texture and lights that follows the mouse gives nice effects😍 The source code is here: #threejsJourney #r3f

Anderson Mancini

47,570 Aufrufe • vor 3 Jahren

Coding agents shouldn't run on a laptop. That's why I've been building Cave. It's a self-hosted platform for running OpenCode agents in isolated sandboxes on your own server. You give it a GitHub repo, it creates an isolated sandbox, clones the code, sets everything up, and gives you a coding agent. Check in from your phone, run multiple agents side by side. It's not open source yet (it will be), but I'm opening a private beta so you can install it on your own server already. I've been dogfooding it since day one. If you're running a software factory: multiple agents working in parallel across repos, Cave gives you one place to monitor and manage all of it. Now, I'm looking for people who want to give it a spin. DM me, happy to set you up 😊 🚀
0:59

Coding agents shouldn't run on a laptop. That's why I've been building Cave. It's a self-hosted platform for running OpenCode agents in isolated sandboxes on your own server. You give it a GitHub repo, it creates an isolated sandbox, clones the code, sets everything up, and gives you a coding agent. Check in from your phone, run multiple agents side by side. It's not open source yet (it will be), but I'm opening a private beta so you can install it on your own server already. I've been dogfooding it since day one. If you're running a software factory: multiple agents working in parallel across repos, Cave gives you one place to monitor and manage all of it. Now, I'm looking for people who want to give it a spin. DM me, happy to set you up 😊 🚀

André König

36,113 Aufrufe • vor 4 Monaten

And now I’m here. Strand on the Green, Chiswick, London. A very pleasant place, with beautiful houses, the River Thames and more importantly, three very lovely pubs. I’m tempted to go in to one.
1:26

And now I’m here. Strand on the Green, Chiswick, London. A very pleasant place, with beautiful houses, the River Thames and more importantly, three very lovely pubs. I’m tempted to go in to one.

Alastair Hilton

51,041 Aufrufe • vor 4 Monaten

Here's a quick way to get an interactive, link-worthy calculator on your website in less than 10 minutes. No plugins or coding experience needed. If I get a few comments with the word "calculator", I'll post the full source code and ready to use template in the thread. 🧵
0:27

Here's a quick way to get an interactive, link-worthy calculator on your website in less than 10 minutes. No plugins or coding experience needed. If I get a few comments with the word "calculator", I'll post the full source code and ready to use template in the thread. 🧵

Sacha | JournoFinder.com

14,585 Aufrufe • vor 3 Jahren

This might be the most precise mechanic in the game, League runs on 30Hz servers and must time it in one server tick (or 1/30th of a second, 0.03s window) or it doesn't work. Combo is: - 3 stack opponent - Path forward - Flip movement and path backwards - Cancel your backwards pathing with R on the exact server tick the enemy skillshot is fired 0.2x Speed vs Normal Speed version below, you can barely see it in normal speed.
0:23

This might be the most precise mechanic in the game, League runs on 30Hz servers and must time it in one server tick (or 1/30th of a second, 0.03s window) or it doesn't work. Combo is: - 3 stack opponent - Path forward - Flip movement and path backwards - Cancel your backwards pathing with R on the exact server tick the enemy skillshot is fired 0.2x Speed vs Normal Speed version below, you can barely see it in normal speed.

No Arm Whatley

44,886 Aufrufe • vor 1 Jahr

Coding just leveled up for me ⚡ Hooked Claude Code into GLM-4.5 → suddenly it’s like pair-programming with a senior dev: 👉 Bootstrapped a full Next.js CRM in minutes 👉 Auto-wired SQLite + Prisma with zero friction 👉 Guided me through git commits & pushes hands-free The kicker? The whole build cost under $1. 🤯 And now GLM Coding Plans launch with promo pricing: • Lite — $3/mo • Pro — $15/mo (40–60% faster + image/video + web search) • MAX — $30/mo (all features unlocked 🚀) Premium coding speed, open-source pricing. This combo might be the future of dev workflows. 👉
0:53

Coding just leveled up for me ⚡ Hooked Claude Code into GLM-4.5 → suddenly it’s like pair-programming with a senior dev: 👉 Bootstrapped a full Next.js CRM in minutes 👉 Auto-wired SQLite + Prisma with zero friction 👉 Guided me through git commits & pushes hands-free The kicker? The whole build cost under $1. 🤯 And now GLM Coding Plans launch with promo pricing: • Lite — $3/mo • Pro — $15/mo (40–60% faster + image/video + web search) • MAX — $30/mo (all features unlocked 🚀) Premium coding speed, open-source pricing. This combo might be the future of dev workflows. 👉

Alif Hossain

17,991 Aufrufe • vor 10 Monaten

Nothing beats open-source! We now have a fully open-source coding agent that can compete with any proprietary solution out there: • 500+ available models • It doesn't train on your code • It doesn't charge a penny for tokens • It's open. You can change if you want. Kilo Code is the #1 app on OpenRouter right now, and the community is releasing new features on GitHub at rocket speed (about 20 releases in the last 30 days). I installed it in Visual Studio Code, but you can install it in pretty much any IDE. I need to spend more time with it, but here is a very interesting design choice: Most coding agents support two modes: an "agentic" mode for writing code and an "ask" mode for talking to the model without making any changes. Kilo Code supports 5 different modes: 1. Architect mode - Research, design, and planning 2. Debug mode - Problem solving and troubleshooting 3. Code mode - To write code. 4. Ask mode - To talk to the model. 5. Orchestrator mode - Combine every agent to solve problems. On paper, I really like this. I'll need more time to try out the specialized agents and see whether niching becomes an advantage. Here is the link to the website: Thanks to the Kilo for partnering with me on this post.
6:17

Nothing beats open-source! We now have a fully open-source coding agent that can compete with any proprietary solution out there: • 500+ available models • It doesn't train on your code • It doesn't charge a penny for tokens • It's open. You can change if you want. Kilo Code is the #1 app on OpenRouter right now, and the community is releasing new features on GitHub at rocket speed (about 20 releases in the last 30 days). I installed it in Visual Studio Code, but you can install it in pretty much any IDE. I need to spend more time with it, but here is a very interesting design choice: Most coding agents support two modes: an "agentic" mode for writing code and an "ask" mode for talking to the model without making any changes. Kilo Code supports 5 different modes: 1. Architect mode - Research, design, and planning 2. Debug mode - Problem solving and troubleshooting 3. Code mode - To write code. 4. Ask mode - To talk to the model. 5. Orchestrator mode - Combine every agent to solve problems. On paper, I really like this. I'll need more time to try out the specialized agents and see whether niching becomes an advantage. Here is the link to the website: Thanks to the Kilo for partnering with me on this post.

Santiago

73,062 Aufrufe • vor 9 Monaten

Atomic Agent beat Hermes on GAIA: 69.8% vs. 58.5%, and finished 1.6× faster! They ran both agents through all 53 GAIA Level 1 tasks using the same 4-bit Qwen-3.6-35B on the same Apple M4 Max. - Atomic: 37/53 solved in 3h 12m - Hermes: 31/53 solved in 5h 10m Atomic solved six more tasks and finished nearly two hours sooner. It’s an Open-Source-Agent-Runtime model under MIT license. Open Source got a huge push! Love to see!
0:23

Atomic Agent beat Hermes on GAIA: 69.8% vs. 58.5%, and finished 1.6× faster! They ran both agents through all 53 GAIA Level 1 tasks using the same 4-bit Qwen-3.6-35B on the same Apple M4 Max. - Atomic: 37/53 solved in 3h 12m - Hermes: 31/53 solved in 5h 10m Atomic solved six more tasks and finished nearly two hours sooner. It’s an Open-Source-Agent-Runtime model under MIT license. Open Source got a huge push! Love to see!

Chubby♨️

54,239 Aufrufe • vor 23 Tagen

Do not fear, FASTER for #AtariSTE is still very much alive and kicking! Just waiting for a few audio assets to come in before I build the final version to submit to Silly Venture 2024 WE on 5th - 8th December. Binary release and source code will follow after the compo. 🚗🏎️
2:14

Do not fear, FASTER for #AtariSTE is still very much alive and kicking! Just waiting for a few audio assets to come in before I build the final version to submit to Silly Venture 2024 WE on 5th - 8th December. Binary release and source code will follow after the compo. 🚗🏎️

Jonathan Thomas @atariste.bsky.social

12,394 Aufrufe • vor 1 Jahr

Here’s a very interesting lesson on armspeed. This player struggled with casting and early release. I worked on speeding her arms up on the way down and moving lead shoulder faster off the chin. Amazing change to the lag and sync in a few swings 🔥🔥🔥👍🏼
1:10

Here’s a very interesting lesson on armspeed. This player struggled with casting and early release. I worked on speeding her arms up on the way down and moving lead shoulder faster off the chin. Amazing change to the lag and sync in a few swings 🔥🔥🔥👍🏼

JonathanYarwood

40,384 Aufrufe • vor 2 Jahren