James Thomas

inference at @databricks

[James Thomas @jamesjthomas]

GLM 5.2 on the effect of increased page size on FlashInfer decode kernel parallelism

Seems reasonable

[Cartesia @cartesia]

Today, we’re excited to release the first step in our mission to build real time multimodal intelligence for every device: Sonic, a blazing fast  (🚀 135ms model latency), lifelike generative voice model and API.

Read https://t.co/kmqpKoR1NA and try Sonic https://t.co/rMnegk14Jl

[James Thomas @jamesjthomas]

@aerinykim @eladgil Seconded, tried it and it worked great: https://t.co/uf12N8RQxS

[James Thomas @jamesjthomas]

@lloydtabb Are you able to repost this?

[James Thomas @jamesjthomas]

@andygrove_io @sc13ts Is there a Python frontend for this? Doesn't seem like support is there yet in https://t.co/iMBshuBASD

[Matei Zaharia @matei_zaharia]

My research group just released Weld 0.2, plus accelerated versions of NumPy and Pandas: https://t.co/uVY4BuRyZV. All of these are available on PiPY for easy installation and testing.

[Shoumik Palkar @sppalkia]

The Weld paper is now on Arxiv: https://t.co/s9tyfB8K81
New interfaces for data intensive applications can give order of magnitude speedups!

[James Thomas @jamesjthomas]

@kcmetrobus got it from the driver on the reverse trip, thanks a lot

[James Thomas @jamesjthomas]

@kcmetrobus left my backpack on bus 48 at 15th and 52nd at 9 pm on 5/30 -- you guys have it? Thanks

[James Thomas @jamesjthomas]

@divarestaurant Are you interested in sponsoring an Indian cultural event at MIT?

[James Thomas @jamesjthomas]

chillin on my machine