Re: [VOTE] The new direction for project Mnemonic

hnique we are going to > >> achieve. > >> > > >> > On Thu, Mar 21, 2024 at 10:45 AM rhythmwind > >> wrote: > >> > > >> > > +1. Agree with Xiaojin's idea. > >> > > > >> > > If mnemonic project could provid a

Re: [VOTE] The new direction for project Mnemonic

>> and >> > > capacity required by ever increasing demand of ML models, it will be >> much >> > > attractive. >> > > >> > > Thanks Zhen >> > > >> > > Original message >> > > Fr

Re: [VOTE] The new direction for project Mnemonic

ve way to meet speed and > > > capacity required by ever increasing demand of ML models, it will be > much > > > attractive. > > > > > > Thanks Zhen > > > > > > ---- Original message ---- > > > From: Gordon > > > Date: 3/

Re: [VOTE] The new direction for project Mnemonic

creasing demand of ML models, it will be much > > attractive. > > > > Thanks Zhen > > > > Original message > > From: Gordon > > Date: 3/21/24 10:16 AM (GMT-08:00) > > To: dev@mnemonic.apache.org, priv...@mnemonic.apache.org > >

Re: [VOTE] The new direction for project Mnemonic

asing demand of ML models, it will be much > attractive. > > Thanks Zhen > > Original message > From: Gordon > Date: 3/21/24 10:16 AM (GMT-08:00) > To: dev@mnemonic.apache.org, priv...@mnemonic.apache.org > Subject: Re: [VOTE] The new direction for

Re: [VOTE] The new direction for project Mnemonic

@mnemonic.apache.org, priv...@mnemonic.apache.org Subject: Re: [VOTE] The new direction for project Mnemonic Thank you Xiaojing and Chenyang.On Tue, Mar 19, 2024 at 8:10 AM Chenyang Li wrote:> Ok sound good to me then>> +1>>      < = >> LESS IS MORE>>> On Mo

Re: [VOTE] The new direction for project Mnemonic

Thank you Xiaojing and Chenyang. On Tue, Mar 19, 2024 at 8:10 AM Chenyang Li wrote: > Ok sound good to me then > > +1 > > < = > > LESS IS MORE > > > On Mon, Mar 18, 2024 at 6:18 PM Gordon wrote: > > > Hi Chenyang, > > > > You are right, how to optimize the ML/AI workload through

Re: [VOTE] The new direction for project Mnemonic

Ok sound good to me then +1 < = > LESS IS MORE On Mon, Mar 18, 2024 at 6:18 PM Gordon wrote: > Hi Chenyang, > > You are right, how to optimize the ML/AI workload through non-volatile > objects is a big challenge. > > I saw some successful efforts on this domain, for example, Apache

Re: [VOTE] The new direction for project Mnemonic

Hi Xiaojing, Thank you for your insights on this. It is valuable for us to refer to those art of status engineering works for the ML/AI domain. and looks very promising to have all AI/ML parameters, dataset in memory storage to further accelerate the data process. Let's further think about

Re: [VOTE] The new direction for project Mnemonic

Hi Chenyang, You are right, how to optimize the ML/AI workload through non-volatile objects is a big challenge. I saw some successful efforts on this domain, for example, Apache Arrow provided optimized dataframe and numpy array to accelerate the data processing, while they do have DLPack

Re: [VOTE] The new direction for project Mnemonic

Quick thoughts :) Extend our Mnemonic interface to more advanced memory for AI. High-speed emerging memories for AI hardware accelerators | Nature Reviews Electrical Engineering How AI Impacts

Re: [VOTE] The new direction for project Mnemonic

+1, thanks. -- Best Regards, Xiaojin Jiao On Mon, Mar 18, 2024 at 9:58 AM Gordon wrote: > Hi Mnemonic contributors and PMCs, > > As the benefits of non-volatile memory technology are very limited on its > own, we may need to consider developing some applications to present it, > such as

Re: [VOTE] The new direction for project Mnemonic

Ml/ai will be interesting to add, but the problem is I don’t see a clear interception from Mnemonic project. Does anybody have any ideas on this? Chenyang < = > LESS IS MORE On Mon, Mar 18, 2024 at 9:58 AM Gordon wrote: > Hi Mnemonic contributors and PMCs, > > As the benefits of

Re: [VOTE] The new direction for project Mnemonic

+1, Thanks. On Mon, Mar 18, 2024 at 9:57 AM Gordon wrote: > Hi Mnemonic contributors and PMCs, > > As the benefits of non-volatile memory technology are very limited on its > own, we may need to consider developing some applications to present it, > such as ML/AI. so I would like to call a