rishi @rishiiyer01

squeezing water out of stone rishiiyer.com Joined January 2017

Tweets

2K
Followers

720
Following

706
Likes

4K

rishi @rishiiyer01

a day ago

most people get around this by drafting at 0 temperature, which tends to separately tank speculation acceptance length (in most cases)

0 0 6 55 0

View Details

simple way to describe decoherence here: lets say a block of 16 tokens has two viable sequences on decoding present in the logit distributions; independent logit sampling from each slot in the block would cause mixing between two viable sequences resulting in an invalid sequence

1 0 7 79 0

View Details

rishi @rishiiyer01

a day ago

very clever result from deepseek on the serialized head, was literally just discussing this idea of defeating diffusion decoherence with an rnn style head with friends today

elie @eliebakouch

a day ago

new inference optimization method by @deepseek_ai with an extremely detailed paper, draft model and framework to train them. results in production for dsv4 lead to +50% for throughput and latency (can go to ~80% for latency, crazy). full explanation of DSpark: it's about

25 132 988 75K 530

1 2 25 3K 15

View Details

rishi @rishiiyer01

a week ago

my personal choice of numerical integrator is rk2. For cigarette I smoke 大重九

0 0 3 210 0

View Details

Alex Rives @alexrives

2 weeks ago

Together with UC Berkeley we are announcing the laser phase plate - a breakthrough in atomic resolution imaging. This is the brightest continuous wave laser in the world, 100 million times the intensity of the surface of the sun. Phase contrast plays an important role in microscopy, but it was thought close to impossible for electron microscopy, where it would require interfering with an electron beam. Holger Mueller and Robert Glaeser proposed exactly this using a standing wave laser. It has taken over 15 years to make this a reality. Biohub partnered with UC Berkeley and Mueller to support this work and to engineer and build the technology. Contrast has been the critical barrier to achieving atomic resolution imaging of the cell. In cryo-electron tomography, a cellular imaging technology that uses electron microscopy, the low contrast makes it impossible to resolve anything but the largest proteins within their cellular context. The laser phase plate removes that barrier. With advances in AI this breakthrough in contrast will start to open up a new frontier in structural biology, that will allow us to see the molecular machines of the cell, and how they assemble into far more complex and dynamic systems, and understand how they work.

91 575 4K 651K 2K

View Details

rishi @rishiiyer01

2 weeks ago

@TheWhiteTower16 i hope the government makes an effigy out of me

0 0 2 49 0

View Details

rishi @rishiiyer01

2 weeks ago

starting an illegal token dealing business where u can use models even if u are international by leveraging my identity

4 0 26 2K 0

View Details

rishi @rishiiyer01

2 weeks ago

at the end of the day im gonna ask my model to collect the data from the internet rephrase it and create the entire codebase to redo the midtrain and posttrain

stochasm @stochasticchasm

2 weeks ago

RSI only feels real in the mundane way it was real when the kimi team said k2 was helping curate training data, accelerating researchers

3 1 39 4K 3

0 0 7 924 0

View Details

rishi @rishiiyer01

2 weeks ago

🖤

llm_enjoyer @LLMenjoyer

2 weeks ago

@Gcabrielclark @ryu0000000001 Also thanks to @rishiiyer01 for many bits of good advice, like lowk he was the one that convinced me to try MoE at last even though way back then I was not a believer. The model also still has residual traces of @fluffykittnmeow's autism even after so much time.

1 0 14 703 0

1 0 10 468 0

View Details

rishi @rishiiyer01

2 weeks ago

@Gcabrielclark u da real art

0 0 2 25 0

View Details

rishi @rishiiyer01

2 weeks ago

@Gcabrielclark is one of the most hardworking people I have ever been around. v inspirational to be around

Zyphra @ZyphraAI

2 weeks ago

Today we're releasing ZONOS2, our next-generation real-time TTS model with high-fidelity voice cloning. ZONOS2 is the most expressive open-source TTS model, released under Apache 2.0 and available on Zyphra Cloud on @AMD. 🧵

22 104 704 332K 673

1 3 10 376 1

View Details

rishi @rishiiyer01

2 weeks ago

@LLMenjoyer u are on huawei npu in my mind

1 0 1 16 0

View Details

rishi @rishiiyer01

2 weeks ago

i was very surprised to see the recent diffusion-gemma use uniform state diffusion instead of MDM. I think there is little advantage to this approach currently in comparison to mdm or single step mdm with smaller blocks. Especially given the current devices we run inference on.

2 2 7 336 0

View Details

rishi @rishiiyer01

2 weeks ago

main thing here is I have seen with my own eyes no drop off (and even increase) in evals from doing single step spec mdm on an moe with decently high acceptance rates.

0 0 2 121 0

View Details

rishi @rishiiyer01

2 weeks ago

that being said their engineering is still top notch as their throughput is strong. But it can always be stronger through codesign with the arch and devices you run on.