Qi Zhou
@dreamingleo89
Followers
12
Following
84
Statuses
20
Joined August 2011
@NeurIPSConf Something’s seriously wrong. There needs to be actions. We cannot just let this pass by.
1
0
2
@convequity Snow/Databricks etc will never be able to catch up with Google and Meta on OSS model quality. The gap may become smaller gradually but the the investment will hardly pay off. However I don’t think they will stop investing because they need a story about AI, especially Databricks.
2
0
2
@soumithchintala @alignment_lab Intuitively it makes sense. However in reality data privacy could be a significant concern for companies, and they often lack the motivation to open-source human feedback. Therefore incentive mechanisms need to be put in place.
0
0
0
@convequity 7B is decent for story telling. There’s one guy ran 7B Llama on his 4GB RAM Raspberry PI and got ~10 tokens/s - It’s definitely possible to run it way faster on latest phones with kernel fusion, memory optimizations etc. But I agree edge is not there yet
I've sucefully runned LLaMA 7B model on my 4GB RAM Raspberry Pi 4. It's super slow about 10sec/token. But it looks we can run powerful cognitive pipelines on a cheap hardware.
0
0
1
@convequity The 7B llama with some optimizations on normal cpus is already able to generate faster than the speed people read. Autoregressive or not doesn’t matter. Not sending personal data including input prompts to cloud is a huge win for privacy. Edge saves cost for service providers.
1
0
0
@bmcclendon I just read this and miss the good old days when I needed to carefully select mp3 songs to put into my 128mb mp3 player.
0
0
0
RT @martin_maas: I'm excited to share what I have been working on at @GoogleAI. Our @ASPLOSConf paper shows how to reduce C++ memory fragme…
0
74
0
Map Reduce - A really simple introduction « Kaushik Sathupadi http://t.co/Q7dz8xNkJh via @_ksat
0
0
0
@BeatsSupport When will there be a powerbeats2 firmware update? The audio lag issue is really frustrating.
1
0
0