bio
NVIDIA CUDA
Joined September 2022
- Tweets1.4K
- Following520
- Followers1.9K
- Likes2.8K
name retweeted
deepseek has given a lot
+ the first* oss reasoning model (the viral launch of R1, jan 2025)
+ MLA
+ DSA
+ MTP (spec-decoding)
+ FlashMLA
+ DeepGEMM
+ DeepEP
+ EPLB
+ 3FS
+ @DeepSeekHarness
+ DeepInfer
and its all open-source.
thank u so much @deepseek_ai for this all 🙏
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
RT @abhibuilds: oss update:
+ 1 PR merged to @KVCache_AI/ Mooncake
+ 1 PR merge to @vllm_project /semantic-router
more on the go... 🚀
This quoted post is unavailable.
name retweeted
created a new artifact on quantizing L2 HiCache for @sgl_project
did 2 experiments which resulted in
> 43.75% less host KV memory per token: 147,456 → 82,944 bytes
> 1.78x L2 capacity at the same host-memory budget: 54,254 → 96,451 tokens
here is link to artifact - abhinandan.one/artifacts/bui…
also wrote an article which is quoted below.
lmk ur thoughts :)
name retweeted
attention is topic which took me most long to understand.
so spent this week to writing on attention, everything i knew from scratch. u can understand & feel it easilt from from first principles.
pls take a read, lmk ur reviews.
name retweeted
oss update:
+ 1 PR merged to @nvidia 's dynamo
+ 1 PR merged to @vllm_project /vllm-ascend
more on the go.... 🚀
name retweeted
Attention is the first step in understanding the models.
But generally, it is not easy to understand it in one definition or directly jumping to QKV & softmax.
So I tried to make it easy and simple to understand it from the first principles intuition.
name retweeted
oss update:
+ 7 PRs merged to @vllm_project /semantic-router
+ 1 PR merged to @KVCache_AI /Mooncake
After getting just completely stuck & sucked out of this account, I am leaving this account.
You may connect with me on my other completely fresh account @abhinandan_ai
This new account is dedicated for learning, journey & skills, NO shitposting.
> co-founded OpenAI,
> led Tesla Autopilot,
> pioneered deep learning education
> left high-profile corporate roles to focus on pure research and building
> teaches the world AI from scratch for free on YouTube > writes code daily,
> doesn't chase titles, money, or media..
YouTube should provide the right-click context menu
to quickly unsubscribe the unwanted subscriptions
If the salary was exactly the same, which setup are you choosing for the rest of your career?
100%100% Remote (Home)
0%Hybrid (Best of both)
0%100% In-Office
0%4-Day In-Office Work/Week
1 votes • Final resultsI am extremely frustated with the network service of the @reliancejio . Name is True5g but speed in always 0, literally 0.
@JioCare its request + warning, if you don't provide for the money which you have already taken, will go to consumer court, be ready to the respond there
as a indian, I have an extremely strong love for the land,
but an extremely strong hate for its system.
As far as theories of mathematics are about reality, they are not certain; And so far they are certain, they are not about reality.
- Guess Who
Hinduism spirituality will keep you away from any depression.
Hare Krishna
Antidepressant users per 1,000 people:
🇮🇳 India: 9
🇱🇻 Latvia: 21
🇷🇺 Russia: 23
🇰🇷 South Korea: 27
🇭🇺 Hungary: 29
🇱🇹 Lithuania: 36
🇨🇷 Costa Rica: 39
🇪🇪 Estonia: 40
🇮🇹 Italy: 44
🇳🇱 Netherlands: 47
🇸🇰 Slovakia: 48
🇹🇷 Turkey: 48
🇫🇷 France: 54
🇱🇺 Luxembourg: 57
🇮🇱 Israel: 57
🇧🇷 Brazil: 58
🇳🇴 Norway: 61
🇩🇪 Germany: 62
🇦🇹 Austria: 63
🇸🇮 Slovenia: 63
🇨🇿 Czechia: 65
🇬🇷 Greece: 70
🇫🇮 Finland: 81
🇧🇪 Belgium: 83
🇩🇰 Denmark: 84
🇨🇱 Chile: 90
🇪🇸 Spain: 92
🇸🇪 Sweden: 105
🇬🇧 UK: 108
🇺🇸 USA: 110
🇦🇺 Australia: 122
🇨🇦 Canada: 130
🇵🇹 Portugal: 139
🇮🇸 Iceland: 161
Source: OECD (2022), NISD
