Ivan Fioravanti ᯅ's banner

Ivan Fioravanti ᯅ

@ivanfioravanti • 22,336 subscribers

GenAI/LLM addicted, Apple MLX, Cloud computing, Kubernetes, Technology Advisor, Investor and Co-Founder & Board Member of CoreView.

Shorts

MLX MiniMax 2.5 running LOCALLY on a single M3 Ultra 512GB! Writing a poem on LLMs at 6bit quantization! 🔥 Let's start some coding, context and distributed tests! Generation: 40.2 tokens-per-sec Peak memory: 186 GB

MLX MiniMax 2.5 running LOCALLY on a single M3 Ultra 512GB! Writing a poem on LLMs at 6bit quantization! 🔥 Let's start some coding, context and distributed tests! Generation: 40.2 tokens-per-sec Peak memory: 186 GB

226,052 次观看

It worked! Hermes Agent + Exo + Qwen3 Coder Next 8bit to create an incredible snake game, with model following 100% specifications passed in prompt! Let's load something bigger now!

It worked! Hermes Agent + Exo + Qwen3 Coder Next 8bit to create an incredible snake game, with model following 100% specifications passed in prompt! Let's load something bigger now!

159,666 次观看

GLM-4.7 Flash by Z.ai running on M3 Ultra using MLX! 4bit at 81 toks/s 🔥 8bit at 64 toks/s 🔥 (video below) Model converted from transformers to MLX on the fly with Codex + GPT-5.2 HIgh!

GLM-4.7 Flash by Z.ai running on M3 Ultra using MLX! 4bit at 81 toks/s 🔥 8bit at 64 toks/s 🔥 (video below) Model converted from transformers to MLX on the fly with Codex + GPT-5.2 HIgh!

75,425 次观看

This is on M3 Ultra; M5 Ultra will be crazy!

This is on M3 Ultra; M5 Ultra will be crazy!

98,227 次观看

First real test on M4 Max 40GPU to transcribe 3 hours of video with mlx_whisper 0.4.1 - M4 Max 2:19 mins (Max Fan no throttling) - M4 Max 2:29 mins (System Fan throttling towards end) GPU freq goes 1.578 --> 952 Here the video (8x) showing throttling kicking in at the end.

First real test on M4 Max 40GPU to transcribe 3 hours of video with mlx_whisper 0.4.1 - M4 Max 2:19 mins (Max Fan no throttling) - M4 Max 2:29 mins (System Fan throttling towards end) GPU freq goes 1.578 --> 952 Here the video (8x) showing throttling kicking in at the end.

216,328 次观看

Qwen 3 0.6B is Ultra powerful if fine-tuned! Here I got 83.5% on classification tasks! Only Gemini 2.5 Pro does better with 85%! Video is in real time on M3 Ultra!

Qwen 3 0.6B is Ultra powerful if fine-tuned! Here I got 83.5% on classification tasks! Only Gemini 2.5 Pro does better with 85%! Video is in real time on M3 Ultra!

143,105 次观看

Another incredible p5js animation created by the new King: Sonnet 3.7 Thinking! It's a kind of magic!

Another incredible p5js animation created by the new King: Sonnet 3.7 Thinking! It's a kind of magic!

137,251 次观看

Eulerian Fluid simulation test! Zero-shot! Opus 4.6 vs GPT-5.3 vs Gemini 3 Deep Think! My personal preference: 🥇 Gemini 3 Deep Think (really strong!) 🥈 Opus 4.6 🥉 GPT 5.3 High

Eulerian Fluid simulation test! Zero-shot! Opus 4.6 vs GPT-5.3 vs Gemini 3 Deep Think! My personal preference: 🥇 Gemini 3 Deep Think (really strong!) 🥈 Opus 4.6 🥉 GPT 5.3 High

38,277 次观看

GLM-4.7 and MiniMax M2.1 side by side, both powered by Claude Code! Prompt below. Trick: one setting file per provider: - claude --settings ~/.claude/minimax_settings.json - claude --settings ~/.claude/zai-settings.json

GLM-4.7 and MiniMax M2.1 side by side, both powered by Claude Code! Prompt below. Trick: one setting file per provider: - claude --settings ~/.claude/minimax_settings.json - claude --settings ~/.claude/zai-settings.json

44,866 次观看

Playing with Qwen3-TTS is and MLX-Audio locally on Mac Studio M3 Ultra 🔥 Amazing model by Qwen and great work by Prince Canuma bringing this magic to MLX! Tuning the voice with instructions feels like magic! Command and prompts to run it are in the video.

Playing with Qwen3-TTS is and MLX-Audio locally on Mac Studio M3 Ultra 🔥 Amazing model by Qwen and great work by Prince Canuma bringing this magic to MLX! Tuning the voice with instructions feels like magic! Command and prompts to run it are in the video.

35,720 次观看

I'm finally entering the Google Gemini world too! I subscribed to Ultra so I could test Deep Think! But I hit a wall immediately 😢

I'm finally entering the Google Gemini world too! I subscribed to Ultra so I could test Deep Think! But I hit a wall immediately 😢

27,606 次观看

Qwen-Image-Lightning on Apple Silicon! M3 Ultra 9 mins --> 42 seconds! 🔥 M4 Max 11mins --> 1:27 mins! 🤩 1st try: MPS + float32 11 mins 2nd try: MPS + bfloat16 9 mins 3rd try: MPS + bfloat16 + Qwen-Image-Lightning LoRA 42 seconds!!! 🤯 Here a video 8x

Qwen-Image-Lightning on Apple Silicon! M3 Ultra 9 mins --> 42 seconds! 🔥 M4 Max 11mins --> 1:27 mins! 🤩 1st try: MPS + float32 11 mins 2nd try: MPS + bfloat16 9 mins 3rd try: MPS + bfloat16 + Qwen-Image-Lightning LoRA 42 seconds!!! 🤯 Here a video 8x

51,001 次观看

Qwen QwQ 32B fp16 on M4 Max and M2 Ultra powered by MLX! M2 Ultra - 10.2 toks/s M4 Max - 7.6 toks/s! "Create an amazing animation using p5js" o1-mini level local model! Note: use temp 0.7-0.75 for optimal results in coding. I did some tests and this

Qwen QwQ 32B fp16 on M4 Max and M2 Ultra powered by MLX! M2 Ultra - 10.2 toks/s M4 Max - 7.6 toks/s! "Create an amazing animation using p5js" o1-mini level local model! Note: use temp 0.7-0.75 for optimal results in coding. I did some tests and this

62,377 次观看

DeepSeek R1 Qwen 7B 4bit M2 Ultra vs M4 Max on Apple MLX 🤫 Let them think... (video 4x in center part) M2 Ultra: 114.9 tokens per sec M4 Max (14"): 88.3 tokens per sec

DeepSeek R1 Qwen 7B 4bit M2 Ultra vs M4 Max on Apple MLX 🤫 Let them think... (video 4x in center part) M2 Ultra: 114.9 tokens per sec M4 Max (14"): 88.3 tokens per sec

59,734 次观看

I love the ToolCall-15 so much, that I'm working on it to add some features: - mlx and lm-studio providers - batch calling - config params to test temp, top-p, top-k, min-p - quantization display in UI model name Here running on M5 Max 128GB 🚀 Thanks stevibe you rock!

I love the ToolCall-15 so much, that I'm working on it to add some features: - mlx and lm-studio providers - batch calling - config params to test temp, top-p, top-k, min-p - quantization display in UI model name Here running on M5 Max 128GB 🚀 Thanks stevibe you rock!

10,916 次观看

🔥Apple MLX first 6bit model is on Hugging Face!🔥 Qwen2.5-Coder-32B-Instruct-6bit! 3bit conversion and test in progress! Video 8x below on M4 Max 40GPU: - Prompt: 38 tokens, 61.731 tokens-per-sec - Generation: 1181 tokens, 16.939 tokens-per-sec - Peak memory: 25.122 GB

🔥Apple MLX first 6bit model is on Hugging Face!🔥 Qwen2.5-Coder-32B-Instruct-6bit! 3bit conversion and test in progress! Video 8x below on M4 Max 40GPU: - Prompt: 38 tokens, 61.731 tokens-per-sec - Generation: 1181 tokens, 16.939 tokens-per-sec - Peak memory: 25.122 GB

46,494 次观看

MLX Step-3.5-Flash! I've reached 45 toks/s! Thanks to Fast-MLX skill by Awni Hannun used withing Codex with GPT 5.2 High! From 13 toks/s v0 to 45 toks/s v2 🚀 but again I bet Tarjei Mandt will do even better 🙌🏻

MLX Step-3.5-Flash! I've reached 45 toks/s! Thanks to Fast-MLX skill by Awni Hannun used withing Codex with GPT 5.2 High! From 13 toks/s v0 to 45 toks/s v2 🚀 but again I bet Tarjei Mandt will do even better 🙌🏻

13,933 次观看

Here Qwen 3 0.6B again on Italian wines 82.5%! The best commercial LLM I tested in the past was scoring 75.6% (Sonnet 3.5)

Here Qwen 3 0.6B again on Italian wines 82.5%! The best commercial LLM I tested in the past was scoring 75.6% (Sonnet 3.5)

25,367 次观看

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

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 次观看 • 5 个月前

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 次观看 • 3 个月前

MLX + OpenCode + Qwen3.5-122B-A10B-4bit on M3 Ultra created a great snake game! Work zero-shot. Video clearly in super fast mode during generation. I generated the prompt using Grok 4.20, it's in the article.

MLX + OpenCode + Qwen3.5-122B-A10B-4bit on M3 Ultra created a great snake game! Work zero-shot. Video clearly in super fast mode during generation. I generated the prompt using Grok 4.20, it's in the article.

Ivan Fioravanti ᯅ

74,659 次观看 • 3 个月前

Web Design Royal Rumble! Claude Code with: - GLM 4.7 (left) - Opus 4.5 (center) - MiniMax M2.1 (right) 🥇 M2.1 fastest 🥈 Opus 4.5 🥉 GLM 4.7 All great, which one is the best? 🤷🏻‍♂️

Web Design Royal Rumble! Claude Code with: - GLM 4.7 (left) - Opus 4.5 (center) - MiniMax M2.1 (right) 🥇 M2.1 fastest 🥈 Opus 4.5 🥉 GLM 4.7 All great, which one is the best? 🤷🏻‍♂️

Ivan Fioravanti ᯅ

91,414 次观看 • 5 个月前

Kimi K2.5 (Kimi CLI) vs MiniMax 2.1 (CC) vs GLM 4.7 (CC). 🔥 Same prompt to create a single-page website for "PHANTOM PROTOCOL" a fictional tactical shooter video game, 0-shot. Spoiler IMO: 🥇 Kimi K2.5 is another league 🥈 MiniMax 2.1 🥉 GLM 4.7

Kimi K2.5 (Kimi CLI) vs MiniMax 2.1 (CC) vs GLM 4.7 (CC). 🔥 Same prompt to create a single-page website for "PHANTOM PROTOCOL" a fictional tactical shooter video game, 0-shot. Spoiler IMO: 🥇 Kimi K2.5 is another league 🥈 MiniMax 2.1 🥉 GLM 4.7

Ivan Fioravanti ᯅ

70,404 次观看 • 4 个月前

MiniMax M2.1 vs Opus 4.5 vs GLM-4.7 built an interactive 3D solar system from scratch. Running on 3 Claude Code in parallel. The winner is: 🥇 GLM-4.7! 🥈 Opus 4.5 (slowest) 🥉 M2.1 (fastest) GLM-4.7 design capabilities are another level vs 4.6!

MiniMax M2.1 vs Opus 4.5 vs GLM-4.7 built an interactive 3D solar system from scratch. Running on 3 Claude Code in parallel. The winner is: 🥇 GLM-4.7! 🥈 Opus 4.5 (slowest) 🥉 M2.1 (fastest) GLM-4.7 design capabilities are another level vs 4.6!

Ivan Fioravanti ᯅ

80,508 次观看 • 5 个月前

Repo Prompt + o3 = Mind Blowing results! Plain video, no edit!

Repo Prompt + o3 = Mind Blowing results! Plain video, no edit!

Ivan Fioravanti ᯅ

179,828 次观看 • 1 年前

OpenCode + MLX + Qwen3.5-397B-A17B-4bit. Video is 8x, but the goal is showing that It works! This is something unimaginable just few months ago. MLX Team is pushing like crazy and M5 Ultra will do the rest 🚀

OpenCode + MLX + Qwen3.5-397B-A17B-4bit. Video is 8x, but the goal is showing that It works! This is something unimaginable just few months ago. MLX Team is pushing like crazy and M5 Ultra will do the rest 🚀

Ivan Fioravanti ᯅ

48,610 次观看 • 3 个月前

I'm sure many already knows: you can use DeepSeek V3 in Cursor right now: - add as OpenAI Base URL - use deepseek/deepseek-chat as model - use it in chat (not in composer) Here the combo created a snake game eating AI companies. 3 turns needed to get it working 1/2 🧵 Sonnet below

I'm sure many already knows: you can use DeepSeek V3 in Cursor right now: - add as OpenAI Base URL - use deepseek/deepseek-chat as model - use it in chat (not in composer) Here the combo created a snake game eating AI companies. 3 turns needed to get it working 1/2 🧵 Sonnet below

Ivan Fioravanti ᯅ

153,359 次观看 • 1 年前

DeepSeek-R1-0528-5bit on MLX pushing M3 Ultra 512GB to its limits! 501GB used mem visibile on mactop in the video! Context: 4K tokens Prompt: 190.29 t/s Gen: 11.37 t/s Peak Mem: 487.48 GB! THIS IS APPLE MLX!

DeepSeek-R1-0528-5bit on MLX pushing M3 Ultra 512GB to its limits! 501GB used mem visibile on mactop in the video! Context: 4K tokens Prompt: 190.29 t/s Gen: 11.37 t/s Peak Mem: 487.48 GB! THIS IS APPLE MLX!

Ivan Fioravanti ᯅ

102,949 次观看 • 11 个月前

OpenCode + GLM-4.7-Flash 8bit. llama-server vs mlx_lm.server. Using one M3 Ultra for each host. Same prompts, you can check duration for each step in the video. Amazing experience on both side! Local coding AI model are becoming reality! 🔥

OpenCode + GLM-4.7-Flash 8bit. llama-server vs mlx_lm.server. Using one M3 Ultra for each host. Same prompts, you can check duration for each step in the video. Amazing experience on both side! Local coding AI model are becoming reality! 🔥

Ivan Fioravanti ᯅ

47,063 次观看 • 4 个月前

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 Altoukhi 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 Altoukhi here!

Ivan Fioravanti ᯅ

14,102 次观看 • 1 个月前

Kimi K2.5 - Same Prompt - OpenCode vs KIMI CLI vs Claude Code, any differences? 🤔 Create a single-page website for "PHANTOM PROTOCOL" a fictional tactical shooter video game. Design capabilities of this model are out of scale! 🥇 Final results at the end of the video. 🧵

Kimi K2.5 - Same Prompt - OpenCode vs KIMI CLI vs Claude Code, any differences? 🤔 Create a single-page website for "PHANTOM PROTOCOL" a fictional tactical shooter video game. Design capabilities of this model are out of scale! 🥇 Final results at the end of the video. 🧵

Ivan Fioravanti ᯅ

44,652 次观看 • 4 个月前

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.

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.

Ivan Fioravanti ᯅ

19,575 次观看 • 1 个月前

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

19,258 次观看 • 1 个月前

🔥 Google Gemini 2 can see and comment your screen and search on internet for additional information! This is sci-fi! WOW! 🤯

🔥 Google Gemini 2 can see and comment your screen and search on internet for additional information! This is sci-fi! WOW! 🤯

Ivan Fioravanti ᯅ

130,485 次观看 • 1 年前

Mistral Vibe CLI running Devstral-small-2-2512 on M3 Ultra 512GB with LM Studio as backend 🔥 First part of the video 1x and the remaining 2x Great speed overall, 27 toks/s as average tracked on LM Studio console. Details to run it below 🧵

Mistral Vibe CLI running Devstral-small-2-2512 on M3 Ultra 512GB with LM Studio as backend 🔥 First part of the video 1x and the remaining 2x Great speed overall, 27 toks/s as average tracked on LM Studio console. Details to run it below 🧵

Ivan Fioravanti ᯅ

49,682 次观看 • 5 个月前

I did it! It works! Using GLM-4.7-4bit with mlx_lm.server and opencode to fix real code locally! 🔥 Here single M3 Ultra 512GB, nex step phase will be 2 using Tensor Parallelism and then apply same changes to exo. Prefill is slow on a single machine, but generation is good.

I did it! It works! Using GLM-4.7-4bit with mlx_lm.server and opencode to fix real code locally! 🔥 Here single M3 Ultra 512GB, nex step phase will be 2 using Tensor Parallelism and then apply same changes to exo. Prefill is slow on a single machine, but generation is good.

Ivan Fioravanti ᯅ

44,000 次观看 • 5 个月前

MiniMax M2.7 in Droid seems stronger than in Claude Code! Here in Lego San Andreas creation based on a prompt by dr bosconovitch I was not expecting such a difference 👀 I need to use Droid more!

MiniMax M2.7 in Droid seems stronger than in Claude Code! Here in Lego San Andreas creation based on a prompt by dr bosconovitch I was not expecting such a difference 👀 I need to use Droid more!

Ivan Fioravanti ᯅ

24,039 次观看 • 2 个月前

Claude Code battle: GLM 4.7 vs MiniMax 2.1 creating fireworks effect. 🎇 I bet 2026 will be an amazing year for AI!

Claude Code battle: GLM 4.7 vs MiniMax 2.1 creating fireworks effect. 🎇 I bet 2026 will be an amazing year for AI!

Ivan Fioravanti ᯅ

39,673 次观看 • 5 个月前