abheecs (@abheecs) header
abheecs (@abheecs) profile picture

Applied AI/ML | AIML Junior

Joined April 2025
Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)Photo by abheecs (@abheecs)
Happy Diwali Folks🪔🪔
1
3
68
You can't improve what you dont measure
1
2
84
Legend
Replying to @lydiahallie
wait is it just me or is Opus 5.5 dumber
3
115
Have been doing some crazy upskilling in the past couple of weeks. Building things that you want to use yourself is the best thing which you can do to upskill yourself honestly… build it use the frontier models break it down reason around it.
1
69
Have been using herdr for the past week. Love it.
2
68
Been writing a lot of Go recently. Love it. Love Opus 5.5 aswell
1
73
The only time you truly fail, is when you decide to give up.
11,750
47,511
7,253
314,151
26,823,675
How to trigger real focus before you study: - Before you start, run a silent script in your head: "I need to learn this" - Tell yourself the information is important to voluntarily ramp up alertness and focus - Don't wait for material to pull your attention out of you - engage it volitionally - When your mind drifts, bring it back, constantly. That is the work - Put on a hoodie and hat as blinders so you can only attend to what's in front of you - Expect strain and agitation. That's the signal, not a failure The strain reflects release of neuromodulators like epinephrine, which cue your circuits that they need to change later. No strain, no neuroplasticity. @hubermanlab
12
150
11
1,376
57,773
Idk what to do in life atp feel lost kinda
1
2
214
Is it just me or when you first read about spec dec on a very high level. It doesn't make a lot of sense.
3
189
Also use an LLM before doing something stupid. Not saying that I did 😊
if you're trying to profile CUDA kernels (nsight compute / ncu) on a serverless GPU platform like Modal it silently won't work
1
3
190
if you're trying to profile CUDA kernels (nsight compute / ncu) on a serverless GPU platform like Modal it silently won't work
1
1
2
359
lfg pt 2
1
2
88
Was building a CUDA agent and kinda need the cheapest GPU rental that wont die mid session any suggestions?
77
Im never giving up. Never.
1
130
abheecs retweeted
MESSI IS THE PLAYER RONALDO WISHES HE WAS.
1,294
5,863
221
36,221
368,379
Really did not know ZeroMQ is this important for SGlang
126
After this you can check out @LiteLLM it's an amazing piece of product. @ishaan_jaff you've solved an interesting problem
99