Video wird geladen...

Video konnte nicht geladen werden

Beim Laden dieses Videos ist ein Problem aufgetreten. Dies könnte an einem vorübergehenden Netzwerkproblem liegen oder das Video ist möglicherweise nicht verfügbar.

MiniMax M3 support added to mlx-vlm with MSA implementation! 🚀 Tested on M3 Ultra 512GB running at 24 tps with peak memory ~240GB. Now working on optimizing performance and adding ton of tests 💪 Model is here: PR is here:

Ivan Fioravanti ᯅ

38,527 subscribers

24,376 Aufrufe • vor 2 Monaten •via X (Twitter)

Gaming Bildung Wissenschaft & Technologie

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

Kimi K2.6 was released 1h ago, and it looks amazing! Here it's running with MLX (mlx-vlm) on two M3 Ultras (full 1T param VLM) 🔥

Kimi K2.6 was released 1h ago, and it looks amazing! Here it's running with MLX (mlx-vlm) on two M3 Ultras (full 1T param VLM) 🔥

Pedro Cuenca

65,768 Aufrufe • vor 4 Monaten

GLM-4.7-8bit (350GB) running at 19 toks/s on two M3 Ultra 512GB using Tensor Parallelism with EXO - MLX, versus 14 toks/s with single node. 🚀 Now context benchmarking & then OpenCode tests 🔥 Note: this is from sources, I had to change things to run it.

GLM-4.7-8bit (350GB) running at 19 toks/s on two M3 Ultra 512GB using Tensor Parallelism with EXO - MLX, versus 14 toks/s with single node. 🚀 Now context benchmarking & then OpenCode tests 🔥 Note: this is from sources, I had to change things to run it.

Ivan Fioravanti ᯅ

327,687 Aufrufe • vor 8 Monaten

MLX GLM 5.2 Distributed on two M3 Ultra 512GB 🔥 One M3 Ultra: 18.8 tokens/sec Two M3 Ultra: 23.4 tokens/sec Context: - PR by Pedro Cuenca is still open and probably there is room for improvement: - basic generation test to measure decoding performance here, I will do a full context benchmarking once PR is more mature - nvfp4 quantization used - Video alternates standard speed and x20, with one Mac first and distributed later. Enjoy! 🙌🏻

MLX GLM 5.2 Distributed on two M3 Ultra 512GB 🔥 One M3 Ultra: 18.8 tokens/sec Two M3 Ultra: 23.4 tokens/sec Context: - PR by Pedro Cuenca is still open and probably there is room for improvement: - basic generation test to measure decoding performance here, I will do a full context benchmarking once PR is more mature - nvfp4 quantization used - Video alternates standard speed and x20, with one Mac first and distributed later. Enjoy! 🙌🏻

Ivan Fioravanti ᯅ

87,805 Aufrufe • vor 2 Monaten

Qwen3.5-397B-A17B with Vision running on a Single M3 Ultra 🚀 Shout out to the Qwen team because they cooked with this model! Get started today: > uv pip install -U mlx-vlm Model collection 👇🏽 Repo:

Qwen3.5-397B-A17B with Vision running on a Single M3 Ultra 🚀 Shout out to the Qwen team because they cooked with this model! Get started today: > uv pip install -U mlx-vlm Model collection 👇🏽 Repo:

Prince Canuma

11,386 Aufrufe • vor 6 Monaten

This is MiniMax-M2.5 MLX running in LM Studio on an Apple Mac Studio M3 Ultra 512GB. Fast enough out of the box for hosting OpenClaw, n8n workflows, and Open WebUI for the team.

This is MiniMax-M2.5 MLX running in LM Studio on an Apple Mac Studio M3 Ultra 512GB. Fast enough out of the box for hosting OpenClaw, n8n workflows, and Open WebUI for the team.

Patrick J Kennedy

76,574 Aufrufe • vor 6 Monaten

What is this Nanbeige4.1-3B model running at - 77 toks/s in bf16 (in video) - 115 toks/s in 8bit on M3 Ultra with MLX with these benchmark scores! 🔥

What is this Nanbeige4.1-3B model running at - 77 toks/s in bf16 (in video) - 115 toks/s in 8bit on M3 Ultra with MLX with these benchmark scores! 🔥

Ivan Fioravanti ᯅ

85,228 Aufrufe • vor 6 Monaten

Running four high-level OpenCode agents + subagents with mlx_lm.server continuous batching and MiniMax M2.5 (6-bit). Fits easily on a 512GB M3 Ultra. Generation is quite fast. But prefill is still slow compared to cloud servers.

Running four high-level OpenCode agents + subagents with mlx_lm.server continuous batching and MiniMax M2.5 (6-bit). Fits easily on a 512GB M3 Ultra. Generation is quite fast. But prefill is still slow compared to cloud servers.

Awni Hannun

25,535 Aufrufe • vor 6 Monaten

First results are in. Llama 4 Maverick 17B active / 400B total is blazing fast with MLX on an M3 Ultra. Here is the 4-bit model generating 1100 tokens at 50 tok/sec:

First results are in. Llama 4 Maverick 17B active / 400B total is blazing fast with MLX on an M3 Ultra. Here is the 4-bit model generating 1100 tokens at 50 tok/sec:

Awni Hannun

149,855 Aufrufe • vor 1 Jahr

Running Ring 1T reasoning model on a single M3 Ultra with mlx-lm. It's quantized to 3.5 bits-per-weight. Uses 440GB and generated ~6k tokens at 18.2 toks/sec. Getting closer to GPT-5 at home.

Running Ring 1T reasoning model on a single M3 Ultra with mlx-lm. It's quantized to 3.5 bits-per-weight. Uses 440GB and generated ~6k tokens at 18.2 toks/sec. Getting closer to GPT-5 at home.

Awni Hannun

55,138 Aufrufe • vor 11 Monaten

Running Minimax M2.1 (MiniMax (official)) with OpenCode (OpenCode) and mlx_lm.server. Works quite well on an M3 Ultra. Once the KV cache is warm the prompt processing is pretty quick. And token generation is very fast.

Running Minimax M2.1 (MiniMax (official)) with OpenCode (OpenCode) and mlx_lm.server. Works quite well on an M3 Ultra. Once the KV cache is warm the prompt processing is pretty quick. And token generation is very fast.

Awni Hannun

32,329 Aufrufe • vor 8 Monaten

DeepSeek-V4-Flash powering 4 parallel agents on Pi (by Mario Zechner) 🚀 Running on M3 Ultra at ~30-34 tok/s and 160-187GB peak URAM using MLX-LM. Special shoutout to clandestine.eth 🦇🔊, Pedro Cuenca, Tarjei Mandt, Ivan Fioravanti ᯅ and others for helping optimize and shape this PR. PR:

DeepSeek-V4-Flash powering 4 parallel agents on Pi (by Mario Zechner) 🚀 Running on M3 Ultra at ~30-34 tok/s and 160-187GB peak URAM using MLX-LM. Special shoutout to clandestine.eth 🦇🔊, Pedro Cuenca, Tarjei Mandt, Ivan Fioravanti ᯅ and others for helping optimize and shape this PR. PR:

Prince Canuma

109,941 Aufrufe • vor 4 Monaten

Open-weight MiniMax M3 filled out a US customs form from a driver's license photo For this test we deployed MiniMax M3 Q4 using MLX-VLM on a Mac Studio M3 Ultra 512GB RAM. The model was tasked with reading a scanned document and an ID card photo, then completing a declaration form Output: 736 tokens · Input: 1,847 tokens · Time: ~31s The model analyzed both inputs, streamed its reasoning, and then called three tools: write_field for text fields, mark for Yes/No checkboxes, and sign for the signature and date. It extracted the required information, mapped it to the correct fields and completed the form without any manual input

Open-weight MiniMax M3 filled out a US customs form from a driver's license photo For this test we deployed MiniMax M3 Q4 using MLX-VLM on a Mac Studio M3 Ultra 512GB RAM. The model was tasked with reading a scanned document and an ID card photo, then completing a declaration form Output: 736 tokens · Input: 1,847 tokens · Time: ~31s The model analyzed both inputs, streamed its reasoning, and then called three tools: write_field for text fields, mark for Yes/No checkboxes, and sign for the signature and date. It extracted the required information, mapped it to the correct fields and completed the form without any manual input

atomic.chat

109,369 Aufrufe • vor 2 Monaten

A long time coming but new mlx-lm is here with better batching support in the server and Gemma 4. pip install -U mlx-lm Here is a video where a single M3 Ultra serves 5 opencode sessions with Gemma 4 26B that process ~130k tokens in ~1.5 minutes.

A long time coming but new mlx-lm is here with better batching support in the server and Gemma 4. pip install -U mlx-lm Here is a video where a single M3 Ultra serves 5 opencode sessions with Gemma 4 26B that process ~130k tokens in ~1.5 minutes.

Angelos Katharopoulos

66,202 Aufrufe • vor 4 Monaten

GLM-5 runs with mlx-lm on a single 512GB M3 Ultra in Q4. It's quite good in my initial testing and pretty fast as well. It generated a highly functional space invaders game using 7.1k tokens at 15.4 tok/s and 419GB memory. Thanks to Gökdeniz Gülmez and Tarjei Mandt for the port.

GLM-5 runs with mlx-lm on a single 512GB M3 Ultra in Q4. It's quite good in my initial testing and pretty fast as well. It generated a highly functional space invaders game using 7.1k tokens at 15.4 tok/s and 419GB memory. Thanks to Gökdeniz Gülmez and Tarjei Mandt for the port.

Awni Hannun

60,599 Aufrufe • vor 6 Monaten

🎉 Congrats to MiniMax (official) on releasing MiniMax M3! Frontier coding and agentic capabilities, native image and video input, computer use, and a 1M-token context window, all in a single open model. At the heart of M3 is MSA, a new sparse attention architecture: instead of attending densely over the full KV cache, each query scores 128-token KV blocks and runs attention only over the top blocks. That is what makes 1M-token context practical to serve. M3 runs in vLLM with day-0 support, verified on NVIDIA and AMD hardware: ✨ MSA sparse attention with dedicated prefill and decode kernels ✨ 1M-token context serving with prefix caching and chunked prefill ✨ BF16 and MXFP8 checkpoints, with MoE backends for both Hopper and Blackwell ✨ Native multimodal input (image + video) ✨ Tool calling, reasoning parsing, and thinking-mode control for agent workloads Day-0 support like this is a true team effort. Grateful to the teams at MiniMax (official), NVIDIA AI, AI at AMD, and Inferact, and to the vLLM community for making it happen. 🙏 Deep dive into the implementation, kernel work, and deployment recipes: 🔗

🎉 Congrats to MiniMax (official) on releasing MiniMax M3! Frontier coding and agentic capabilities, native image and video input, computer use, and a 1M-token context window, all in a single open model. At the heart of M3 is MSA, a new sparse attention architecture: instead of attending densely over the full KV cache, each query scores 128-token KV blocks and runs attention only over the top blocks. That is what makes 1M-token context practical to serve. M3 runs in vLLM with day-0 support, verified on NVIDIA and AMD hardware: ✨ MSA sparse attention with dedicated prefill and decode kernels ✨ 1M-token context serving with prefix caching and chunked prefill ✨ BF16 and MXFP8 checkpoints, with MoE backends for both Hopper and Blackwell ✨ Native multimodal input (image + video) ✨ Tool calling, reasoning parsing, and thinking-mode control for agent workloads Day-0 support like this is a true team effort. Grateful to the teams at MiniMax (official), NVIDIA AI, AI at AMD, and Inferact, and to the vLLM community for making it happen. 🙏 Deep dive into the implementation, kernel work, and deployment recipes: 🔗

vLLM

40,306 Aufrufe • vor 2 Monaten

GLM-4.7 runs quite well on an M3 Ultra with mlx-lm, even at a near lossless precision (6-bit here). It generated the best space invaders game I've seen yet for a local model (even included sound effects!). Generated 6600 tokens and ran at 16 tok/s.

GLM-4.7 runs quite well on an M3 Ultra with mlx-lm, even at a near lossless precision (6-bit here). It generated the best space invaders game I've seen yet for a local model (even included sound effects!). Generated 6600 tokens and ran at 16 tok/s.

Awni Hannun

237,191 Aufrufe • vor 8 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!

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

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!

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 Aufrufe • vor 4 Monaten

Linear scaling achieved with multiple DeepSeek v3.1 instances. 4x macs = 4x throughput. 2x M3 Ultra Mac Studios = 1x DeepSeek @ 14 tok/sec 4x M3 Ultra Mac Studios = 2x DeepSeek @ 28 tok/sec DeepSeek V3.1 is a 671B parameter model - so at its native 8-bit quantization, it requires ~700GB of memory to run the model. EXO puts half of the layers on each device, combining their memory. EXO uses MLX distributed with TB5 interconnect, optimized for Apple Silicon. If we need higher throughput, adding two more devices lets us serve more users at once. EXO Labs handles all of this seamlessly - adding more devices to the cluster for linear scaling as we need it. The new EXO 1.0 will be open-source soonTM

Linear scaling achieved with multiple DeepSeek v3.1 instances. 4x macs = 4x throughput. 2x M3 Ultra Mac Studios = 1x DeepSeek @ 14 tok/sec 4x M3 Ultra Mac Studios = 2x DeepSeek @ 28 tok/sec DeepSeek V3.1 is a 671B parameter model - so at its native 8-bit quantization, it requires ~700GB of memory to run the model. EXO puts half of the layers on each device, combining their memory. EXO uses MLX distributed with TB5 interconnect, optimized for Apple Silicon. If we need higher throughput, adding two more devices lets us serve more users at once. EXO Labs handles all of this seamlessly - adding more devices to the cluster for linear scaling as we need it. The new EXO 1.0 will be open-source soonTM

Matt Beton

158,485 Aufrufe • vor 1 Jahr

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! 💪

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 4 Monaten