Загрузка видео...

Не удалось загрузить видео

На главную

Google’s Gemma 4 E2B running on-device on iPhone 17 Pro Gemma 4 is built from the same research as Gemini 3, has image understanding capabilities and can reason if needed Running at ~40tk/s with MLX optimized for Apple Silicon

adrgrondin's profile picture

Adrien Grondin

8,274 subscribers

1,044,880 просмотров • 4 месяцев назад •via X (Twitter)

Здоровье и велнесОбразованиеНаука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

Google's Gemma 4 26B A4B running on MacBook Pro M5 Max 100% native Swift Achieving over 100tk/s with Apple MLX framework
0:14

Google's Gemma 4 26B A4B running on MacBook Pro M5 Max 100% native Swift Achieving over 100tk/s with Apple MLX framework

Adrien Grondin

351,153 просмотров • 4 месяцев назад

Demo of 1-bit Bonsai 8B from PrismML running on-device on iPhone 17 Pro More than 40tk/s for a dense 8B model on iPhone, that’s a first Powered by Apple MLX and available now in Locally AI
0:15

Demo of 1-bit Bonsai 8B from PrismML running on-device on iPhone 17 Pro More than 40tk/s for a dense 8B model on iPhone, that’s a first Powered by Apple MLX and available now in Locally AI

Adrien Grondin

160,895 просмотров • 4 месяцев назад

Apple wasn’t kidding, the iPhone 17 Pro is really built for running LLMs Here’s LFM2 8B A1B by running on-device with MLX in Locally AI - Local AI Chat, the iPhone runs the 8B model with zero struggle Thanks Prince Canuma for the port to MLX, it made the MLX Swift port possible
0:17

Apple wasn’t kidding, the iPhone 17 Pro is really built for running LLMs Here’s LFM2 8B A1B by running on-device with MLX in Locally AI - Local AI Chat, the iPhone runs the 8B model with zero struggle Thanks Prince Canuma for the port to MLX, it made the MLX Swift port possible

Adrien Grondin

493,561 просмотров • 10 месяцев назад

Here’s Apple’s foundation model running on iPhone 17 Pro It’s just so fast Apple was not joking, the A19 Pro chip is really good for running LLMs on-device
0:11

Here’s Apple’s foundation model running on iPhone 17 Pro It’s just so fast Apple was not joking, the A19 Pro chip is really good for running LLMs on-device

Adrien Grondin

1,565,834 просмотров • 11 месяцев назад

Claude fable 5 is cooked Claude fable 5 Vs Gemma 4 26b a4b qat MTP Gemma 4 26b running locally on my 8GB vram single RTX 4060 built this using three.js in a single session and 3 prompts. no cloud, no subscription 100% private unlimited use. how long until it can catch up completely?
0:19

Claude fable 5 is cooked Claude fable 5 Vs Gemma 4 26b a4b qat MTP Gemma 4 26b running locally on my 8GB vram single RTX 4060 built this using three.js in a single session and 3 prompts. no cloud, no subscription 100% private unlimited use. how long until it can catch up completely?

Alok

30,802 просмотров • 2 месяцев назад

Meet Gemma 3n, a model that runs on as little as 2GB of RAM 🤯 It shares the same architecture as Gemini Nano, and is engineered for incredible performance. We added audio understanding, so now it’s multimodal, fast and lean, and runs on-device (no cloud connection required!)
0:15

Meet Gemma 3n, a model that runs on as little as 2GB of RAM 🤯 It shares the same architecture as Gemini Nano, and is engineered for incredible performance. We added audio understanding, so now it’s multimodal, fast and lean, and runs on-device (no cloud connection required!)

Google AI

228,822 просмотров • 1 год назад

Gemma 4 analyzes the video. Generates key questions. Calls Falcon Perception. "Find all the people." 156 found. "Detect only white cars." 8 found. A 26B model is running agentic multi-QA vision orchestration. The models are running locally on a MacBook with MLX. No API.
0:16

Gemma 4 analyzes the video. Generates key questions. Calls Falcon Perception. "Find all the people." 156 found. "Detect only white cars." 8 found. A 26B model is running agentic multi-QA vision orchestration. The models are running locally on a MacBook with MLX. No API.

Maziyar PANAHI

158,954 просмотров • 4 месяцев назад

Another demo of the iPhone 17 Pro’s on-device LLM performance This time with Ling mini 2.0 by InclusionAI, a 16B MoE model with 1.4B active parameters running at ~120tk/s Thanks to Awni Hannun for the MLX DWQ 2-bit quants
0:11

Another demo of the iPhone 17 Pro’s on-device LLM performance This time with Ling mini 2.0 by InclusionAI, a 16B MoE model with 1.4B active parameters running at ~120tk/s Thanks to Awni Hannun for the MLX DWQ 2-bit quants

Adrien Grondin

46,205 просмотров • 11 месяцев назад

Gemma 4 just dropped. I had it captioning video in real-time within an hour. Running locally on a MacBook. No cloud. No API. Real-time scene understanding. Oh and SAM3 is segmenting every object in the same frame. Same laptop.
0:16

Gemma 4 just dropped. I had it captioning video in real-time within an hour. Running locally on a MacBook. No cloud. No API. Real-time scene understanding. Oh and SAM3 is segmenting every object in the same frame. Same laptop.

Maziyar PANAHI

196,846 просмотров • 4 месяцев назад

Introducing MLX-Swift-TS An SDK for running time series foundation models fully on-device on Apple Silicon. When I joined Datadog, Inc. , I was introduced to Toto, our time series foundation model, and got excited about zero-shot forecasting across different domains. While building a health copilot app, I realized there wasn’t a simple way to run models like these locally on device. So I built one. MLX-Swift-TS exposes a common TimeSeriesForecaster interface for loading and running multiple time series architectures directly in Swift using MLX. No server required. The attached video shows on-device forecasting running inside a native Swift app. Huge thanks to Awni Hannun and the MLX team for building MLX and its Swift API, Prince Canuma for inspiration on MLX SDK patterns, and Ameet Talwalkar and the Datadog team for Toto.
0:17

Introducing MLX-Swift-TS An SDK for running time series foundation models fully on-device on Apple Silicon. When I joined Datadog, Inc. , I was introduced to Toto, our time series foundation model, and got excited about zero-shot forecasting across different domains. While building a health copilot app, I realized there wasn’t a simple way to run models like these locally on device. So I built one. MLX-Swift-TS exposes a common TimeSeriesForecaster interface for loading and running multiple time series architectures directly in Swift using MLX. No server required. The attached video shows on-device forecasting running inside a native Swift app. Huge thanks to Awni Hannun and the MLX team for building MLX and its Swift API, Prince Canuma for inspiration on MLX SDK patterns, and Ameet Talwalkar and the Datadog team for Toto.

Kunal Batra

12,585 просмотров • 5 месяцев назад

Gemma 4 is here! Our most intelligent open models to date, are built on the same world-class research and tech as Gemini 3, and are sized to run and fine-tune efficiently on local hardware. Check out what Google Gemma 4 brings to devs: 💎 Advanced Reasoning: Deep logic tasks, complex multi-step planning, and beyond 💎 Longer context: Seamlessly analyze entire codebases with context windows of 128K tokens for our edge models and 256K tokens for our largest models 💎 Vision and audio: Rich, multimodal interactions out of the box 💎 140+ languages: Trained on 140+ languages 💎 Apache 2.0 license: industry-standard open-source license
0:19

Gemma 4 is here! Our most intelligent open models to date, are built on the same world-class research and tech as Gemini 3, and are sized to run and fine-tune efficiently on local hardware. Check out what Google Gemma 4 brings to devs: 💎 Advanced Reasoning: Deep logic tasks, complex multi-step planning, and beyond 💎 Longer context: Seamlessly analyze entire codebases with context windows of 128K tokens for our edge models and 256K tokens for our largest models 💎 Vision and audio: Rich, multimodal interactions out of the box 💎 140+ languages: Trained on 140+ languages 💎 Apache 2.0 license: industry-standard open-source license

Google for Developers

269,912 просмотров • 4 месяцев назад

How much faster is the new MacBook Pro for AI inference? M4 Max is 27% faster with 72 tok/sec compared to 56 tok/sec of the M3 Max with MLX running Gemma 2 9B (4bit). The 27% speedup is the same with Llama-3.2-1b, Llama-3.2-3b and others. Next up: EXO Labs M4 cluster.
0:15

How much faster is the new MacBook Pro for AI inference? M4 Max is 27% faster with 72 tok/sec compared to 56 tok/sec of the M3 Max with MLX running Gemma 2 9B (4bit). The 27% speedup is the same with Llama-3.2-1b, Llama-3.2-3b and others. Next up: EXO Labs M4 cluster.

Alex Cheema

529,690 просмотров • 1 год назад

Gemma 4 looks at a parking lot. Decides what to ask. Calls SAM 3.1. "Segment all vehicles." 64 found. "Now just the white ones." 23 found. One model reasoning and orchestrating. One model executing. Both running locally on a MacBook. MLX. No cloud. No API.
0:14

Gemma 4 looks at a parking lot. Decides what to ask. Calls SAM 3.1. "Segment all vehicles." 64 found. "Now just the white ones." 23 found. One model reasoning and orchestrating. One model executing. Both running locally on a MacBook. MLX. No cloud. No API.

Maziyar PANAHI

594,212 просмотров • 4 месяцев назад

Run Gemma 4 26b MTP on 8 GB VRAM GPUs at 25+ tokens/second. Flags included! local llm space is moving at terminal velocity. only 3 days ago google released gemma 4 26b a4b qat quants. more efficient than before, ran on 8gb vram at 20 tok/sec. and now just a few hours ago, mainline llama.cpp merged a massive update and we just shattered our own record. decode throughput went 25-40% up on the same 8 GB VRAM setup! Before MTP: 20 tps -> After MTP: 28 tps! llama.cpp just officially merged PR #23398 ("add Gemma4 MTP"), bringing native Multi-Token Prediction (MTP) support to Gemma 4 models. By running speculative drafting on the same 8GB VRAM RTX 4060 setup, my decode throughput on a 64k context instantly leaped to a blistering 25–27 tokens/sec thats 25-30% increase with the same hardware. Here is the architectural catch you need to know: Unlike the Qwen 3.5 and 3.6 series, which bake the MTP heads directly into the base GGUF, the Gemma 4 MTP head is not built in. You must download a separate, specialized MTP drafter GGUF (the assistant model) to act as the speculator. (I've dropped the download link in the replies). copy and try the exact flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp --spec-draft-n-max 6 --spec-draft-p-min 0.7 --spec-draft-model gemma-4-26b-A4B-it-assistant-Q4_0.gguf -c 64000 -v n-max 4 and p-min 0.7 is also worth checking out. benchmark on your setup and workflow. if you have a single 8 gb vram nvidia rtx 4060, 3060, 3070, 2080, 2070, grab the MTP drafter GGUF link in the comments and try it yourself. Check it out even if you have asmaller or a larger gpu, such as a single rtx 3090, 4090, 3060, 2060. MTP works for all gemma 4 sizes such as gemma 4 12b, gemma 4 31b etc. but remember to grab the correct mtp draft assistant models respectively. what are you benchmarking today
0:15

Run Gemma 4 26b MTP on 8 GB VRAM GPUs at 25+ tokens/second. Flags included! local llm space is moving at terminal velocity. only 3 days ago google released gemma 4 26b a4b qat quants. more efficient than before, ran on 8gb vram at 20 tok/sec. and now just a few hours ago, mainline llama.cpp merged a massive update and we just shattered our own record. decode throughput went 25-40% up on the same 8 GB VRAM setup! Before MTP: 20 tps -> After MTP: 28 tps! llama.cpp just officially merged PR #23398 ("add Gemma4 MTP"), bringing native Multi-Token Prediction (MTP) support to Gemma 4 models. By running speculative drafting on the same 8GB VRAM RTX 4060 setup, my decode throughput on a 64k context instantly leaped to a blistering 25–27 tokens/sec thats 25-30% increase with the same hardware. Here is the architectural catch you need to know: Unlike the Qwen 3.5 and 3.6 series, which bake the MTP heads directly into the base GGUF, the Gemma 4 MTP head is not built in. You must download a separate, specialized MTP drafter GGUF (the assistant model) to act as the speculator. (I've dropped the download link in the replies). copy and try the exact flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp --spec-draft-n-max 6 --spec-draft-p-min 0.7 --spec-draft-model gemma-4-26b-A4B-it-assistant-Q4_0.gguf -c 64000 -v n-max 4 and p-min 0.7 is also worth checking out. benchmark on your setup and workflow. if you have a single 8 gb vram nvidia rtx 4060, 3060, 3070, 2080, 2070, grab the MTP drafter GGUF link in the comments and try it yourself. Check it out even if you have asmaller or a larger gpu, such as a single rtx 3090, 4090, 3060, 2060. MTP works for all gemma 4 sizes such as gemma 4 12b, gemma 4 31b etc. but remember to grab the correct mtp draft assistant models respectively. what are you benchmarking today

Alok

200,913 просмотров • 2 месяцев назад

Wdym it stopped charging in a properly chilled AC room with the ceiling fan running at full speed 😭 Stuff like this is why I will never consider an iPhone as my primary device (atleast until they fix this). (And this is the iPhone 17 Pro, the one with a vapour chamber 🥀)
0:12

Wdym it stopped charging in a properly chilled AC room with the ceiling fan running at full speed 😭 Stuff like this is why I will never consider an iPhone as my primary device (atleast until they fix this). (And this is the iPhone 17 Pro, the one with a vapour chamber 🥀)

Aryan Gupta

143,407 просмотров • 2 месяцев назад

Iceman has now spent 27 consecutive days at #1 on Apple Music USA, and is 1 day from entering the top 17 longest running #1 albums of all time on the platform in the US🧊
0:19

Iceman has now spent 27 consecutive days at #1 on Apple Music USA, and is 1 day from entering the top 17 longest running #1 albums of all time on the platform in the US🧊

Akademiks TV

18,100 просмотров • 2 месяцев назад

gemma-4-12B-agentic-fable5-composer2.5 V2 is out. the agentic upgrade to the model trained on Fable 5's reasoning. Running it now with TurboQuant llama.cpp on a single RTX 4060( 8 GB VRAM) at 30 tokens/second with full 25000 context and reasoning: # The benchmarks v2 is built for coding + agentic work. writing code, running commands, using tools, debugging, multi step technical tasks. The clearest signal is tau2 bench telecom, an agentic tool use benchmark whose diagnose → fix → verify loop mirrors real terminal/debugging work: tau2 bench telecom numbers: base Gemma 4 12B: ~15% this finetune: ~55%. (Self reported) thats a huge jump # TheTom/llama-cpp-turboquant flags: llama-server.exe -m gemma4-v2-Q4_K_M.gguf -ngl 99 -c 25000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 Flag breakdown: -ngl 99 → full GPU offload -c 25000 → 25K context --cache-type-k q8_0 --cache-type-v turbo3 → mixed-precision KV cache — K at 8-bit, V at ~3-bit via TurboQuant (Walsh Hadamard rotated polar quant, Google's own KV-compression research). Not even merged into mainline llama.cpp. running it off a fork. No API. No cloud. Just llama.cpp. well, a fork of it and any 6gb+ GPU. If you tried yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF, check this out and share your experience with the models
0:16

gemma-4-12B-agentic-fable5-composer2.5 V2 is out. the agentic upgrade to the model trained on Fable 5's reasoning. Running it now with TurboQuant llama.cpp on a single RTX 4060( 8 GB VRAM) at 30 tokens/second with full 25000 context and reasoning: # The benchmarks v2 is built for coding + agentic work. writing code, running commands, using tools, debugging, multi step technical tasks. The clearest signal is tau2 bench telecom, an agentic tool use benchmark whose diagnose → fix → verify loop mirrors real terminal/debugging work: tau2 bench telecom numbers: base Gemma 4 12B: ~15% this finetune: ~55%. (Self reported) thats a huge jump # TheTom/llama-cpp-turboquant flags: llama-server.exe -m gemma4-v2-Q4_K_M.gguf -ngl 99 -c 25000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 Flag breakdown: -ngl 99 → full GPU offload -c 25000 → 25K context --cache-type-k q8_0 --cache-type-v turbo3 → mixed-precision KV cache — K at 8-bit, V at ~3-bit via TurboQuant (Walsh Hadamard rotated polar quant, Google's own KV-compression research). Not even merged into mainline llama.cpp. running it off a fork. No API. No cloud. Just llama.cpp. well, a fork of it and any 6gb+ GPU. If you tried yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF, check this out and share your experience with the models

Alok

145,913 просмотров • 2 месяцев назад

GEMMA 4 26B ON AN RTX 4060 WITH A 248K TOKEN CONTEXT WINDOW 20 tokens per second and a context window so large you can feed it entire codebases, books and research papers in a single prompt this is not a cloud api and not a server rack, this is a regular consumer gpu running locally with llama.cpp and q4_k_xl quantization 248k context on an 8gb vram card was not supposed to be possible and here it is just running on someone’s desk the article below covers exactly which tools and configs make this kind of setup work in 2026 ↓
0:19

GEMMA 4 26B ON AN RTX 4060 WITH A 248K TOKEN CONTEXT WINDOW 20 tokens per second and a context window so large you can feed it entire codebases, books and research papers in a single prompt this is not a cloud api and not a server rack, this is a regular consumer gpu running locally with llama.cpp and q4_k_xl quantization 248k context on an 8gb vram card was not supposed to be possible and here it is just running on someone’s desk the article below covers exactly which tools and configs make this kind of setup work in 2026 ↓

leopardracer

56,772 просмотров • 2 месяцев назад

Gemma 4 watches raw video. Understands the scene. Then prompts SAM 3 to segment and RF-DETR to track. One AI directing two others. Fighter jets. Crowds. Aerial defense footage. All three models running locally on a MacBook. No cloud. What scene should I point this at next?
0:16

Gemma 4 watches raw video. Understands the scene. Then prompts SAM 3 to segment and RF-DETR to track. One AI directing two others. Fighter jets. Crowds. Aerial defense footage. All three models running locally on a MacBook. No cloud. What scene should I point this at next?

Maziyar PANAHI

370,294 просмотров • 4 месяцев назад

Llama.cpp running on a Sun Ultra 45 from 2006. Dual 1.6GHz UltraSPARC IIIs kick out ~0.7 tok/s on Gemma 3 270M Q8_0. Haven’t beat the compiler’s scalar code yet as the UltraSPARC’s VIS vector stuff isn’t ideal for int8 dotprod. May find some perf with Q4_0 nibble unpack. Thanks to Georgi Gerganov for a pretty portable codebase and for whomever saved me the trouble of handling big endian with the s390x port lol
0:19

Llama.cpp running on a Sun Ultra 45 from 2006. Dual 1.6GHz UltraSPARC IIIs kick out ~0.7 tok/s on Gemma 3 270M Q8_0. Haven’t beat the compiler’s scalar code yet as the UltraSPARC’s VIS vector stuff isn’t ideal for int8 dotprod. May find some perf with Q4_0 nibble unpack. Thanks to Georgi Gerganov for a pretty portable codebase and for whomever saved me the trouble of handling big endian with the s390x port lol

Nathan Odle

70,576 просмотров • 3 месяцев назад