Numan
6.5K posts
Numan
@numandev1
Agentic Ai | Loop Engineer | Mobile Application 🕸️ Stackoverflow 25k reputation 👨💻Author of 🗜️react-native-compressor🔑🔑🔑react-native-keys
Keswick, United Kingdom Katılım Ağustos 2020
589 Takip Edilen199 Takipçiler
Apple's M7 Ultra chip coming in 2029 is rumored to support 1.5TB of RAM.
This would make the processor much more capable for on-device AI.
English
@numandev1 I’m playing at lot with cuBLAS
* cuBLASLt
* cuDNN
* CUDA Runtime
* NPP (image processing)
* NVRTC
English
It depends on the user. I think the recent Mac price increase has pushed the DGX Spark into the spotlight. I’m leveraging CUDA libraries more and more, so for me it’s absolutely worth it.
You can’t get a system with CUDA, 128 GB of unified memory, and a 4 TB SSD for around $4,000. The memory bandwidth isn’t the highest, but there’s so much you can do locally with the CUDA ecosystem and 128 GB of memory.
English
@ExileAI_0 @u1tra_instinct are you using 8-bit quantized version for glm 5.2?
English
@numandev1 Several actually but i will point you to @u1tra_instinct and he can provide details as well as a list of others that are as well.
English
he may get actual spark 😂 protect him from wires
jon@lursor_lover
New limited edition spark just dropped
English
@numandev1 goal is to finetune models. I am experimenting with it but nothing solid. I am using open models too.
English
Command Code is #3 on OpenRouter.
We just put the load of a single model there. With our full load we'd be doing a lot better.
#3 in Productivity and Personal Agents
#2 in Programming App & #1 in Native App Builders
Ahmad Awais@MrAhmadAwais
update: Command Code is 2nd most used coding agent now, LFG!!
English
@MrAhmadAwais @0xSero you are right spark is great for building and improving models but for my multimodal orchestration use case its 128 gb memory is limiting
English
@numandev1 @0xSero mac doesn't compare to blackwell gpus — and when you're building/improving models, blackwell gpus are great.
English
@0xSero Why Mac Studio? When you got two Sparks. Also planning to buy sparks. Do they get hot?
English
@MrAhmadAwais @0xSero why I am saying it that I am using this mac from two year which is pretty much good but it is not good to test multiple agent loop with orchestration
English
@MrAhmadAwais @0xSero dont buy it, it have only 128 unified memory, wait for more memory
English
Numan retweetledi
How Expensify uses Agent Device for mobile bug evidence and profiling ⬇️💯
Callstack Engineers@callstackio
Agents reason about code well, but they lack proper tooling to see results of code changes running live on a device. Integrating agent-device into the development loop gives the agent hands and eyes on devices and simulators, so they can reproduce bugs, record the proof, measure performance, and profile renders on its own. We wrote about how Expensify uses agent-device to do that. Read the article here 👇
English
Numan retweetledi
driver.js is now at 1M+ weekly downloads
felt like a good time to give it a refresh
English
Numan retweetledi
LegendList 3.2.0 has a new experimental_adaptiveRender prop, which can improve blanking if you have heavy components. You can render light components when scrolling over a velocity threshold. So you could skip images, gesture detectors, modal setup, etc...
Right has it enabled.
English
@MrAhmadAwais which server you are using to deploy GLM 5.2 model? i am planning to deploy it for my usage
English