IBM Granite: A Family of Open Foundation Models for Code Intelligence • Hacker News Board

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#21

Post by Jedd » Wed, May 08, 2024, 3:43 AM UTC

hustwindmaple1 wrote: post #6

I wonder why companies like IBM are jumping on the LLM bandwagon and training/releasing models that have no chance of competing with Llama/Mistral? To me it just looks like a complete waste of $$ because nobody will use them in any serious scenarios

> ... models that have no chance of competing with ...

I've not seen any proper evaluations for Granite against, say, Llama or Mistral.

Until we do it's probably too early to say they can't compete, at least in some areas where others perform poorly.

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#22

Post by morgante » Wed, May 08, 2024, 3:47 AM UTC

rolisz wrote: post #17

_lvbh wrote: post #16

Earlier quoted context omitted.

Llama and Mistral are already local & fulfill these requirements

Yes, but nobody got fired for buying IBM

That's a pretty outdated phrase, even in enterprise.

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#23

Post by jazzyjackson » Wed, May 08, 2024, 4:10 AM UTC

_lvbh wrote: post #16

xarope wrote: post #15

Earlier quoted context omitted.

Enterprises think differently. They want data provenance, privacy, ability to mitigate/transfer risk etc. If IBM is willing to offer that, there will be enterprises that bite.

Llama and Mistral are already local & fulfill these requirements

but who can you pay to run these models and fulfill these requirements /for you/ ;)

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#24

Post by szszrk » Wed, May 08, 2024, 4:12 AM UTC

antod wrote: post #20

rolisz wrote: post #17

Earlier quoted context omitted.

Yes, but nobody got fired for buying IBM

Yeah if something you install doesn't work, you get the blame. If IBM supplies something that doesn't work (likely), you get to blame them instead.

Between "works" and "doesn't work" there is a full rainbow of possible answers, KPIs, yearly reviews in a network of matrix reporting.

There will be market for their services. Maybe a different one, but there will be.

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#25

Post by andsoitis » Wed, May 08, 2024, 4:34 AM UTC

propter_hoc wrote: post #14

mhh__ wrote: post #12

Earlier quoted context omitted.

IBM do a mixture of shovelware and extremely hardcore tech so they could honestly go either way with this.

> mixture of shovelware and extremely hardcore tech Citation needed All I've seen from them in my professional experience is actually legacy mainframe maintenance.. Not shovelware, but very far from hardcore tech.

“The South Korean technology giant Samsung Electronics was awarded a total of 6,165 United States patents in 2023, the most of any company. Qualcomm ranked second among companies, with 3,854 U.S. patents granted, followed by the likes of Taiwan Semiconductor Manufacturing Company and IBM.” — https://www.statista.com/statistics/274825/companies-with-th...

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#26

Post by esafak » Wed, May 08, 2024, 5:11 AM UTC

propter_hoc wrote: post #14

mhh__ wrote: post #12

Earlier quoted context omitted.

IBM do a mixture of shovelware and extremely hardcore tech so they could honestly go either way with this.

> mixture of shovelware and extremely hardcore tech Citation needed All I've seen from them in my professional experience is actually legacy mainframe maintenance.. Not shovelware, but very far from hardcore tech.

IBM Research is hardcore.

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#27

Post by semi-extrinsic » Wed, May 08, 2024, 5:13 AM UTC

mhh__ wrote: post #12

hustwindmaple1 wrote: post #6

I wonder why companies like IBM are jumping on the LLM bandwagon and training/releasing models that have no chance of competing with Llama/Mistral? To me it just looks like a complete waste of $$ because nobody will use them in any serious scenarios

IBM do a mixture of shovelware and extremely hardcore tech so they could honestly go either way with this.

Agreed. For example their research lab in Zurich has been absolutely world-leading in things like atomic force microscopy (AFM) for four decades, including the Nobel prize in Physics in 1986 (AFM) and 1987 (high-temperature superconductivity). They also invented things like trellis coding and token ring.

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#28

Post by abdullin » Wed, May 08, 2024, 5:21 AM UTC

Jedd wrote: post #21

hustwindmaple1 wrote: post #6

I wonder why companies like IBM are jumping on the LLM bandwagon and training/releasing models that have no chance of competing with Llama/Mistral? To me it just looks like a complete waste of $$ because nobody will use them in any serious scenarios

> ... models that have no chance of competing with ... I've not seen any proper evaluations for Granite against, say, Llama or Mistral. Until we do it's probably too early to say they can't compete, at least in some areas where others perform poorly.

They are Ok-ish.

Previous Granite models were on the level of first llama in my benchmarks.

I’m expecting this version to be roughly comparable to llama 2

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#29

Post by abdullin » Wed, May 08, 2024, 5:34 AM UTC

_lvbh wrote: post #16

xarope wrote: post #15

Earlier quoted context omitted.

Enterprises think differently. They want data provenance, privacy, ability to mitigate/transfer risk etc. If IBM is willing to offer that, there will be enterprises that bite.

Llama and Mistral are already local & fulfill these requirements

IBM goes at great lengths to train models on clean data that has lower risk of copyright or legal issues attached. Just take a look at the model description.

That data issue is important enough for some companies to pick mediocre model over llama or mistral.

Re: IBM Granite: A Family of Open Foundation Models for Code Intelligence

#30

Post by logicchains » Wed, May 08, 2024, 5:55 AM UTC

hustwindmaple1 wrote: post #6

I wonder why companies like IBM are jumping on the LLM bandwagon and training/releasing models that have no chance of competing with Llama/Mistral? To me it just looks like a complete waste of $$ because nobody will use them in any serious scenarios

>I wonder why companies like IBM are jumping on the LLM bandwagon and training/releasing models that have no chance of competing with Llama/Mistral

Did you even read the benchmarks they post on that link? Assuming they're not outright lying, their 8B model is superior to Llama/Mistral models of the same size for coding tasks.