Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

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,...

MattBeton's profile picture

Matt Beton

2,570 subscribers

158,485 Aufrufe • vor 10 Monaten •via X (Twitter)

Wissenschaft & TechnologieBildung
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

NVIDIA sent us 2 DGX Sparks. For a while we wondered what we would do with them. The memory bandwidth is 273GB/s making it 3x slower than an M3 Ultra (819GB/s) for batch_size=1 inference. But it has 4x more FLOPS (100 TFLOPS compared to 26 TFLOPS). So we thought, what if we could combine the DGX Spark & M3 Ultra, and make use of both the massive compute on the DGX Spark and the massive memory-bandwidth on the M3 Ultra. We came up with a way to split inference across both devices and achieve a speedup of up to 4x for long prompts compared to the M3 Ultra on its own. Full details in the blog post linked below.
1:13

NVIDIA sent us 2 DGX Sparks. For a while we wondered what we would do with them. The memory bandwidth is 273GB/s making it 3x slower than an M3 Ultra (819GB/s) for batch_size=1 inference. But it has 4x more FLOPS (100 TFLOPS compared to 26 TFLOPS). So we thought, what if we could combine the DGX Spark & M3 Ultra, and make use of both the massive compute on the DGX Spark and the massive memory-bandwidth on the M3 Ultra. We came up with a way to split inference across both devices and achieve a speedup of up to 4x for long prompts compared to the M3 Ultra on its own. Full details in the blog post linked below.

Alex Cheema

281,225 Aufrufe • vor 8 Monaten

Running GLM-4.7-Flash on 4 x M4 Pro Mac Minis using EXO Labs. Uses tensor parallelism with RDMA over Thunderbolt & MLX backend (h/t Awni Hannun). Runs at 100 tok/sec. We're working on optimizing this at EXO Labs. Aiming to hit ~200 tok/sec on this setup soon.
0:22

Running GLM-4.7-Flash on 4 x M4 Pro Mac Minis using EXO Labs. Uses tensor parallelism with RDMA over Thunderbolt & MLX backend (h/t Awni Hannun). Runs at 100 tok/sec. We're working on optimizing this at EXO Labs. Aiming to hit ~200 tok/sec on this setup soon.

Alex Cheema

62,144 Aufrufe • vor 5 Monaten

🔥DeepSeek-v3-0324-8bit (~700GB) running on 2 x M3 Ultra 512GB ~12 tokens/sec Here the video!
0:25

🔥DeepSeek-v3-0324-8bit (~700GB) running on 2 x M3 Ultra 512GB ~12 tokens/sec Here the video!

Ivan Fioravanti ᯅ

19,761 Aufrufe • vor 1 Jahr

Running GLM-4.7-Flash with OpenCode locally on M4 Max MacBook Pro. 4-bit model runs at 82 tok/sec. Prefill will get ~4x faster with M5 Max MacBook Pro (~28 Jan). EXO will also support disaggregating prefill and decode across devices, e.g. DGX Spark.
0:43

Running GLM-4.7-Flash with OpenCode locally on M4 Max MacBook Pro. 4-bit model runs at 82 tok/sec. Prefill will get ~4x faster with M5 Max MacBook Pro (~28 Jan). EXO will also support disaggregating prefill and decode across devices, e.g. DGX Spark.

Alex Cheema

127,902 Aufrufe • vor 5 Monaten

"Mac Minis for example are a very good fit" - Andrej Karpathy Andrej Karpathy shouted out my work on EXO Labs in his keynote at Y Combinator AI SUS! Here's the breakdown: Right now most AI workloads run in the cloud where requests from different users are continuously batched together. These workloads are FLOPS-bound and favors hardware with the best unit economics of $ per FLOP, i.e. enterprise GPUs. The personal computing revolution will shift these workloads to personal devices with lower batch sizes (mostly batch_size=1). batch_size=1 inference is memory-bound, because all of the model parameters need to be loaded into the GPU every time a token is generated. Apple Silicon with its Unified Memory architecture has a lot of memory and memory bandwidth per $ compared to other hardware: - M4 Pro Mac Mini, 24GB @ 273GB/s, $58.33/GB, $5.13/GB/s - H100, 80GB @ 3350GB/s, $625/GB, $14.93/GB/s The unit economics of Apple Silicon are becoming more compelling with every release of the Mac. The future of AI inference looks more like open weights models (OpenAI open weights model soonTM) run at low batch_size on personal devices.
0:20

"Mac Minis for example are a very good fit" - Andrej Karpathy Andrej Karpathy shouted out my work on EXO Labs in his keynote at Y Combinator AI SUS! Here's the breakdown: Right now most AI workloads run in the cloud where requests from different users are continuously batched together. These workloads are FLOPS-bound and favors hardware with the best unit economics of $ per FLOP, i.e. enterprise GPUs. The personal computing revolution will shift these workloads to personal devices with lower batch sizes (mostly batch_size=1). batch_size=1 inference is memory-bound, because all of the model parameters need to be loaded into the GPU every time a token is generated. Apple Silicon with its Unified Memory architecture has a lot of memory and memory bandwidth per $ compared to other hardware: - M4 Pro Mac Mini, 24GB @ 273GB/s, $58.33/GB, $5.13/GB/s - H100, 80GB @ 3350GB/s, $625/GB, $14.93/GB/s The unit economics of Apple Silicon are becoming more compelling with every release of the Mac. The future of AI inference looks more like open weights models (OpenAI open weights model soonTM) run at low batch_size on personal devices.

Alex Cheema

103,487 Aufrufe • vor 1 Jahr

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! 🙌🏻
0:35

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

87,160 Aufrufe • vor 11 Tagen

M4 Mac AI Coding Cluster Uses EXO Labs to run LLMs (here Qwen 2.5 Coder 32B at 18 tok/sec) distributed across 4 M4 Mac Minis (Thunderbolt 5 80Gbps) and a MacBook Pro M4 Max. Local alternative to Cursor (benchmark comparison soon).
0:20

M4 Mac AI Coding Cluster Uses EXO Labs to run LLMs (here Qwen 2.5 Coder 32B at 18 tok/sec) distributed across 4 M4 Mac Minis (Thunderbolt 5 80Gbps) and a MacBook Pro M4 Max. Local alternative to Cursor (benchmark comparison soon).

Alex Cheema

517,325 Aufrufe • vor 1 Jahr

While we in India should be cautious with the DeepSeek app, we can totally make use of the open source model namesake, if securely deployed on Indian servers, to leapfrog our own AI progress. कृत्रिम has deployed DeepSeek-R1 671B on H100s - first time anywhere in the world. It’s the most powerful open-source model available and to enable all Indian developers to access it, we will price it at ₹1/million tokens for February.
0:46

While we in India should be cautious with the DeepSeek app, we can totally make use of the open source model namesake, if securely deployed on Indian servers, to leapfrog our own AI progress. कृत्रिम has deployed DeepSeek-R1 671B on H100s - first time anywhere in the world. It’s the most powerful open-source model available and to enable all Indian developers to access it, we will price it at ₹1/million tokens for February.

Bhavish Aggarwal

108,596 Aufrufe • vor 1 Jahr

Fuck it, 685B parameter, DeepSeek V3 0324 running locally on M3 Ultra, fully private 🔥 Powered by llama.cpp & dynamic quants from Unsloth AI ⚡ Step 1: brew install llama.cpp Step 2: llama-cli -hf unsloth/DeepSeek-V3-0324-GGUF:Q2_K_XL That's it! 🤗 Honestly a bit surreal to be able to chat with such a chunky model at the touch of the keyboard - future is going to be wild!!
5:00

Fuck it, 685B parameter, DeepSeek V3 0324 running locally on M3 Ultra, fully private 🔥 Powered by llama.cpp & dynamic quants from Unsloth AI ⚡ Step 1: brew install llama.cpp Step 2: llama-cli -hf unsloth/DeepSeek-V3-0324-GGUF:Q2_K_XL That's it! 🤗 Honestly a bit surreal to be able to chat with such a chunky model at the touch of the keyboard - future is going to be wild!!

Vaibhav (VB) Srivastav

168,141 Aufrufe • vor 1 Jahr

For even higher throughput and lower latency: batch generation + tensor parallel with mlx-lm + and mlx.distributed. Here it's generating at 63 tok/sec (throughput) with GLM 4.7 in 6-bit and batch size 4 on 4 M3 Ultras:
0:41

For even higher throughput and lower latency: batch generation + tensor parallel with mlx-lm + and mlx.distributed. Here it's generating at 63 tok/sec (throughput) with GLM 4.7 in 6-bit and batch size 4 on 4 M3 Ultras:

Awni Hannun

22,721 Aufrufe • vor 6 Monaten

Open Source alternative to Apple Intelligence For small tasks, it runs LLM agents on-device. For larger tasks, it runs on your own AI cluster of personal devices Your data stays on your devices By exo intern
1:44

Open Source alternative to Apple Intelligence For small tasks, it runs LLM agents on-device. For larger tasks, it runs on your own AI cluster of personal devices Your data stays on your devices By exo intern

Mohamed Baioumy

142,780 Aufrufe • vor 2 Jahren

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:
0:20

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

Update: Qwen3.5:9b head-to-head (MLX, Apple Silicon optimized) Mac Studio M2 Ultra: 89.74 tok/s Mac Mini M4: 20.82 tok/s MLX basically doubles the speed on both machines.
1:02

Update: Qwen3.5:9b head-to-head (MLX, Apple Silicon optimized) Mac Studio M2 Ultra: 89.74 tok/s Mac Mini M4: 20.82 tok/s MLX basically doubles the speed on both machines.

stevibe

134,256 Aufrufe • vor 4 Monaten

DeepSeek-R1-0528 is now live on Hyperbolic’s Serverless Inference! We also are the first to serve the latest DeepSeek model on Hugging Face. 🟣 Run it instantly:
0:20

DeepSeek-R1-0528 is now live on Hyperbolic’s Serverless Inference! We also are the first to serve the latest DeepSeek model on Hugging Face. 🟣 Run it instantly:

Hyperbolic

33,228 Aufrufe • vor 1 Jahr

Alibaba’s Qwen 3.5 is now running fully on-device on the iPhone 17 Pro. It outperforms models 4x its size, delivers strong visual understanding, and lets you switch reasoning on or off. This demo uses the 2B 6-bit version, optimized with MLX for Apple Silicon.
0:24

Alibaba’s Qwen 3.5 is now running fully on-device on the iPhone 17 Pro. It outperforms models 4x its size, delivers strong visual understanding, and lets you switch reasoning on or off. This demo uses the 2B 6-bit version, optimized with MLX for Apple Silicon.

Hugging Models

202,723 Aufrufe • vor 3 Monaten

A perfect coding model for MLX on Apple silicon.. Qwen delivered again. Runs quite fast on an M3 Ultra. Running the 4-bit quantized with mlx-lm:
0:50

A perfect coding model for MLX on Apple silicon.. Qwen delivered again. Runs quite fast on an M3 Ultra. Running the 4-bit quantized with mlx-lm:

Awni Hannun

186,641 Aufrufe • vor 11 Monaten

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:
0:24

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

24,376 Aufrufe • vor 19 Tagen

I didn't expect DeepSeek v4 PRO (not Flash) to run well on the Mac Studio M3 Ultra with 512GB of RAM. This is 2 bit quantized with the same DwarfStar recipe used for Flash. 433GB GGUF file. 130 t/s prefill, 13 t/s generation. Prefill in the video is low because small prompt.
1:24

I didn't expect DeepSeek v4 PRO (not Flash) to run well on the Mac Studio M3 Ultra with 512GB of RAM. This is 2 bit quantized with the same DwarfStar recipe used for Flash. 433GB GGUF file. 130 t/s prefill, 13 t/s generation. Prefill in the video is low because small prompt.

antirez

169,629 Aufrufe • vor 1 Monat

DeepSeek R1 671B running on 2 M2 Ultras faster than reading speed. Getting close to open-source O1, at home, on consumer hardware. With mlx.distributed and mlx-lm, 3-bit quantization (~4 bpw)
0:54

DeepSeek R1 671B running on 2 M2 Ultras faster than reading speed. Getting close to open-source O1, at home, on consumer hardware. With mlx.distributed and mlx-lm, 3-bit quantization (~4 bpw)

Awni Hannun

865,941 Aufrufe • vor 1 Jahr

MiniMax M2.1 in 4-bit cruises on an M3 Ultra with mlx-lm. Generated a space invaders game using 5098 tokens at 47.2 tok/sec:
0:23

MiniMax M2.1 in 4-bit cruises on an M3 Ultra with mlx-lm. Generated a space invaders game using 5098 tokens at 47.2 tok/sec:

Awni Hannun

93,826 Aufrufe • vor 6 Monaten