~ cooking ~ (@malikkhalidejaz) header
~ cooking ~ (@malikkhalidejaz) profile picture

feeler · thinker · writer #pro_oppressed #divergent Moral 🧭: Respect integrity not authority, humility not delusional superiority, contribution not extortion.

Ontario, Canada
Joined May 2010
Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)Photo by ~ cooking ~ (@malikkhalidejaz)
LLAMA-4G — inference at 4GB VRAM, coming soon on GitHub! With an exciting new harness built-in.
Hypothesis: AI inference engines leave too much on the table. Experiment: Custom inference engine, LLAMA-4G. Not CUDA, yes vulkan. Rig: 4GB VRAM laptop (Quadro T2000) & 32GB RAM (ddr4-2600) running Qwen-3.6-ud-q4_k_xl at 100k context (k8v4) at 380 t/s prefill and 22 t/s decode.
1
453
Hy4 flash is pretty awesome.
5
Knowledge should be free.
2
😂😂😂
> Healthcare in 🇬🇧: 109 year wait time > Healthcare in 🇺🇸: That'll be a million dollars > Healthcare in 🇨🇦:
58
Router was old. Needed a reboot. That’s all.
4
White House is full of sh*t
24
😂
Your email might be down, but the skies are always open. ✉️✈️
2
Scope creep & divergent perfectionism turning an experiment into a product 🙃
LLAMA-4G — inference at 4GB VRAM, coming soon on GitHub! With an exciting new harness built-in.
6
~ cooking ~ retweeted
LibreOffice actually works 365 days a year. Just saying.
1,280
12,079
1,221
93,896
1,784,490
When a racist white Christian learns that Jesus was a brown guy from the Middle East, do they abandon racism, or abandon Christ?
🚨 BREAKING: Sean Hannity announces that he is abandoning his faith because the Pope doesn’t support Trump’s Iran war: “I no longer consider myself a Catholic.”
12
Do people have the freedom to express their dissatisfaction like this, in Iran, to Khamenei’s face?
OMG - This is humiliating. The boos were so loud that he couldn’t hear himself speak. And a woman yelled “traitor” right to his face: trump #maga #usa #politics
13
This is the best one so far 😂
Cry More Orange Pig | New LEGO Video from AI IRAN
14
All whites should be deported back to Europe 🙃
13
LLAMA-4G — runs Qwen-35B on 4GB VRAM Alpha version sneak peek
Getting ready for Qwen4-35B on 4GB VRAM laptops Performance of Qwen-3.6-35B: 400 tok/sec prefill 25 tok/sec decode Vision support added. MTP work in progress.
1
55
No loss in quality, no compromise in numerics.
20
Getting ready for Qwen4-35B on 4GB VRAM laptops Performance of Qwen-3.6-35B: 400 tok/sec prefill 25 tok/sec decode Vision support added. MTP work in progress.
LLAMA-4G — inference at 4GB VRAM, coming soon on GitHub! With an exciting new harness built-in.
1
1
454
Hypothesis: AI inference engines leave too much on the table. Experiment: Custom inference engine, LLAMA-4G. Not CUDA, yes vulkan. Rig: 4GB VRAM laptop (Quadro T2000) & 32GB RAM (ddr4-2600) running Qwen-3.6-ud-q4_k_xl at 100k context (k8v4) at 380 t/s prefill and 22 t/s decode.
2
1
82
MiniCPM and unsloth-iq4 run faster, and at 180k context.
18