My first local AI recipe is out!
Inkling-2-Small on 4× DGX Spark, using Thinking Machines’ original checkpoint weights, with no additional weight quantization. Built on TensorFold with DSpark speculative decoding.
My measurements:
• Single stream with DSpark: 70 tok/s on tool calls, 61 on code, 56 on thinking (31 plain)
• 4 concurrent streams: ~101 tok/s combined at the default settings
• First token at 2K: 1.3 s cold, ~0.2 s cached
• Context up to 1M tokens
github.com/swbiggart/inkling…
Thanks to @thinkymachines for the weights, and @ashxhart + the @tensorfold_dev contributors for the engine.
Would love independent runs and feedback!
5
5
192
Also, thanks to @MiaAI_lab, @bertholomusai, @christopherowen and @OxSero. Seeing the work you’ve been sharing has taught me a lot and pushed me to publish my own recipe.
2
2
29
Scott Biggart retweeted
Great post about the API & Data Portability economy by @Dagk - very kind words about @Uber_API. @woolsey quote too techcrunch.com/2015/06/23/th…
Scott Biggart retweeted
When the Internet works together... teapawt.org/techcrunch/status/6134…
That's me :) "@Uber_SF: Need a last-minute idea for a #BaytoBreakers outfit? This guy has it all figured out. teapawt.org/Uber_SF/status/5993691…"
6
Scott Biggart retweeted
Stock up on plants & fun @ Goettingen Plant Sale in the Portola now @Woolsey
I just offset my @uber use of 2538.47 miles with @TravelGreenApp! #earthday2015 @uber_api @Carbonfundorg travelgreen.io
Scott Biggart retweeted
Don't miss @woolsey's talk on Uber's architecture and tech stack! 6:00pm in the Media Room youtu.be/4zjeIeGtZCs #HackMIT
Today I took a taxi, the El, a plane, a bus, a train, and an uber between Chicago and DC. I love transit!
3
Hey @iamexec why is zipcode 94103 outside your coverage area for cleaning? I'm in downtown SF...
2
A little bit disappointed in @fitbit ‘s “earth aware” packaging for the charger I ordered. twitpic.com/ctx707
1





