@qubitium

GPT-QModel, QvQ ultra low bit quants. Contributor to stuff you are probably using like sglang, vllm, transformers. ucla alum

 Earth
github.com/ModelCloud/
Joined February 2020
  • Tweets7.7K
  • Following4.6K
  • Followers1.7K
881Media
You need to have extreme curiousity if you want humans to sustain this war with AI. Either out monkey the AI, or retire. That's my motto. I am not retiring, yet. Gonna out-curious the Ai until that day comes. Curiosity is my #1 weapon. Not lemming is my #2.
7
1,529
First it was the 10m ctx, then 1b, next came T1000.
Behnam Neyshabur
@bneyshabur2h
Introducing Fern, Mirendil’s first model: pretrained from scratch, with up to 10M tokens of context! That’s ~0.5M lines of code (roughly Megatron-LM scale) or days of an agent’s work, all in context at once. Its first version, Fern 0.1, had a short round of post-training 🚢
104
Can someone reach out to me in dm so I can rent for a few days bare metal Mi350/MI355X? I need near bare metal, not fully virtualized or at least allow me to customize hw virtualization. @tensorwave @CoreWeave @AIatAMD 🙏
3
1
1
401
DeepSeek v4.1 Flash 1017 has never rejected any of my requests due to safety wall and I have used billions tokens on it over many weeks. Curious. What exactly are you people prompting? 😅
2
220
PR: Fix FlyDSL source builds can produce Python bindings tagged for the wrong ABI @AIatAMD
1
3
281
github.com/ROCm/FlyDSL/pull/…

[Bugfix] Pin both MLIR Python detection passes to one interpreter by Qubitium · Pull Request #1328...

Summary Pin both of MLIR's Python detection passes to one interpreter, so the Python bindings are always built for the ABI that will load them. Motivation Building from source could produce...

github.com
2
52
ROCm 10.1 pip install doesn't actually install everything. Need to run "rocm-sdk init" cli 🫠 to unpack the headers/libs. This is anti-pip and bad dev usability. They should've implemented --no-build-isolation and do the unpacking during pip setup ! @AIatAMD Want me to pr?
1
3
672
Btw, most agents miss this and if you ask agent to install, they may always complain about missing lib/headers after "proper" pip install. It's documented by AMD to run rocm-skd init but just because it's documented doesn't' mean it's a good install flow.
64
One word: heat. Whether that is heat from the hbm packaging on/off die, system cooling, I have no clue but I bet my 2 peanuts on heat. AMD has been chipleting long before Nvidia and MI450 is finally showing it's potential end-game to Nvidia semi-monopoly.
Jukan
@jukan0518h
Holy smoke NVIDIA has officially lowered the HBM4 pin speed. 11.7-> 9.3Gbps
1
1
18
1,367
I am not impressed with the public speed results (agentx) for MI350/MI355X for DeepSeek v4.1 Flash. Maybe I should do something about it.
3
222
Frame this image for eternity. I think history will be very kind to them.
Ursula von der Leyen
@vonderleyen12h
Congratulations to my friend @LulaOficial on his electoral victory. Democracy and multilateralism have won in Brazil. Europe’s partnership with Brazil is thriving. We will remain by your side for the years to come. Together we will keep working closely for free and open trade. A rules-based global order. And strong ambitions for our climate.
125
We need to prepare for 400 decode t/s as normal api in 2 months time.
1
2
254
Almost every AI agent/provider/endpoint is dog slow today. I went as far as to test several "fast" endpoints on OpenRouter and they also got slower over the past 12 hours. Either everyone's out of compute, backed up QoS queue or my ip is throttled.
1
181
Documentation is code, it is oracle. Technical writers sometimes needs to be taken out to the sheds. Nasdaq listed hw co still can't find itself to align cpu0/1 as 0 offset based or 1 offset based in its own docs. These two pdf diagrams are back to back next to each other.🫠
1
6
460
I also get a headache reading these docs because almost all hw companies would randomly flip the motherboard 180 degrees in various diagrams. Hello! My head can't spin around like an alien.
1
89
Many Python programs using env variables for flow control/config is a festering disease that should be cured. Just my two cents.
169
🥳 GPT-QModel 7.7.0 dropped! I appreciate the continued contributions from @IntelAI and @positron_ai 👊 * VTPQ added. * Extensive CPU kernels optimized for Zen5 @AIatAMD * Lots of micro optimizations in the quantization pipeline * Many more model support added/clean-ups.
1
1
4
335
github.com/ModelCloud/GPTQMo…

Release GPTQModel v7.7.0 · ModelCloud/GPTQModel

GPTQModel 7.7.0 🚀 CPU inference: Added the shared AVX512-BF16 int4_qzero_group_scaled_cpu kernel for GPTQ/AWQ group sizes 16/32/64/128/256/512 and prioritized it ahead of ATen when supported. Adde...

github.com
96
🥰 Got this dm. Feels good to be acknowledged to have contributed to the advancement of ai in however tiny ways. Many startups, from m to b $ in vc using GPT-QModel,mostly in secret. I wish they upstream their own modifications but competition is so fierce, many don't. 😎
1
4
381
Load more