Building something
London
Joined March 2009
- Tweets2K
- Following740
- Followers420
- Likes11.3K
The one graph going up in the wake of the AI stocks.
Local AI FTW
Remember scalability? We've just created an AI inference engine that scales, and we open-sourced it.
Finally AI running on Apple Silicon not only competes with DGX Sparks, it leaves them in the dust - for now.
github.com/Incept5-AI/inferd
Looking forward to testing @Alibaba_Qwen Qwen4 on this later this week!
We are live on @thursdai_pod with inferd
We wrote an inference engine for running agents, agent-swarms and agentic AI on MacOS.
Finally you can get Nvidia performance from your Apple Mac.
inferd.incept5.ai
1,487 tok/s on Qwen3.8-Flash-Next - Open Source!
Ade Risidore retweeted
We are live on @thursdai_pod with inferd
We wrote an inference engine for running agents, agent-swarms and agentic AI on MacOS.
Finally you can get Nvidia performance from your Apple Mac.
inferd.incept5.ai
1,487 tok/s on Qwen3.8-Flash-Next - Open Source!
Ade Risidore retweeted
Last night, @higgsfield sent @ComfyUI a cease and desist (attached)
They want the Open Source Higgsfield challenge taken down within 48 hours.
Their argument is that our use of the Higgsfield name and that using their CEO's likeness is an infringement violation (man, the irony). Claiming that it makes the event look like Higgsfield is associated with it.
This couldn't be farther from the truth, which is that Higgsfield made a promise to open source its product and never did.
For the past two weeks, builders have been recreating popular closed apps as open workflows anyone can run.
So we asked our community to deliver where Higgsfield failed: apps in the open, built by the community, for the community. We've made it clear that the challenge is organized by ComfyUI and not affiliated with or endorsed by Higgsfield.
When we started Comfy, we made the community a promise. Our product stays open, free, and yours. We’ve kept it every day since, and one letter isn’t going to change that.
The challenge stays up. Comfy and open source will win!
Ade Risidore retweeted
Qwen4, coming soon…
CTO quote was “next week”
Xudong Guo - Research Scientist, Qwen, Token Foundry
Live from Amsterdam
Ade Risidore retweeted
From the @Alibaba_Qwen Cloud CTO, Dr. Feifei Li - “Qwen4 next week”.
Live from Apsara, Amsterdam.
When you consider the short space of time that @Alibaba_Qwen has built this open source ecosystem it’s pretty amazing.
Been playing with an agent on my DGX spark using Qwen3.8-27B nvfp4 and Krea 2 Turbo bf16 and no loras to see how close the base model can get to image replication.
Reference image, basic prompt, score it, rewrite the prompt, repeat.
Started at "a skeleton holding a flask" (0.28), ended up at 0.65 by round 4. 9 rounds, 73 min, all local on the DGX Spark.
Slow yes, easier ways yes, but equally fun watching it evolve.
Reference frame from The Nightmare Before Christmas, all rights belong to Disney and Tim Burton.
Ade Risidore retweeted
FLUX 3 Image Klein is here.
Open Weights with the quality of a large frontier model. Customizable and fast.
5.6B parameters and runs on a single consumer GPU.
Native 4K, bounding boxes, and pixel-perfect editing.
Turned my hand to ACE Step 1.5 today, as an experiment. Spun up a container on my DGX spark then started hitting it.
Turbo is fast, really fast for a quick test, outputs in under 30 seconds.
XL-sft is good, 200 step in under 5 mins for 230s generation.
@ivanfioravanti @jtdavies spin up a container and start playing, it's pretty mind blowing.
this was 200 steps quick tongue in cheek song about the heat in the UK - 5 mins later - Boom! Some minor language adherence issues but still.
Ade Risidore retweeted
Today we're shipping our biggest MLX-VLM release yet: v0.6.0
...and we are raising 💸
This one's about turning your Apple devices into real local agent machines. From your desk to your pocket.
What's new:
⚡ Speculative decoding everywhere — Gemma 4 EAGLE3 + DFlash, Qwen MTP, DeepSeek V4 MTP. Faster tokens, less waiting.
🤖 Agent-ready server — native Anthropic /v1/messages API, stateful /v1/responses, tool calls, Codex context budgets. Plug Claude Code & Codex straight into local models.
👁️ New models galore — DeepSeek V4, ZAYA1-VL, MiniCPM-V 4.6, LFM2 MoE, Step-3.7 Flash, Laguna + more.
🎨 Image gen & editing — FLUX.2 (base + klein), PrismML Bonsai.
🔊 Audio in — Qwen3 Omni, Gemma 4 audio, base64 chat audio.
🧮 TurboQuant KV cache — RHT-correct fast paths for leaner memory.
📦 Modular server, better metrics, cleaner streaming.
Run real agents on the hardware already in your hands.
Github: github.com/Blaizzy/mlx-vlm
huge thanks to @jtdavies for ongoing encouragement, @Prince_Canuma for everything MLX and @pollenrobotics for creating this pocket sized bot for the community to experiment with
