Technology

59438 readers

4276 users here now

This is a most excellent place for technology news and articles.

Our Rules

Follow the lemmy.world rules.
Only tech related content.
Be excellent to each another!
Mod approved content bots can post up to 10 articles per day.
Threads asking for personal tech support may be deleted.
Politics threads may be removed.
No memes allowed as posts, OK to post as comments.
Only approved bots from the list below, to ask if your bot can be added please contact us.
Check for duplicates before posting, duplicates may be removed

Approved Bots

founded 1 year ago

MODERATORS

Gemma 2 is now available to researchers and developers (blog.google)

submitted 4 months ago by simple@lemm.ee to c/technology@lemmy.world

10 comments fedilink hide all child comments

Context: Gemma is a group of free-to-use AI models with a focus on being small. According to benchmarks this outperforms Llama 3.

top 10 comments

sorted by: hot top controversial new old

[–] ExtravagantEnzyme@lemm.ee 14 points 4 months ago (1 children)

Just to be clear, Gemma is only partially open sourced in select area's of the code.

[–] eager_eagle@lemmy.world 4 points 4 months ago (1 children)

how are the weights partially open?

[–] ExtravagantEnzyme@lemm.ee 2 points 4 months ago (1 children)

Only portions of the code are published while the rest is kept under wraps. Classic corporate America bs finding a loop hole to use a trendy term.

[–] eager_eagle@lemmy.world 2 points 4 months ago* (last edited 4 months ago) (1 children)

neural network weights are just files, collections of numbers forming matrices; how is a partially open collection of weights of any use

the weights are open

$ docker exec -it ollama ollama show gemma:7b
  Model                              
  	arch            	gemma	             
  	parameters      	9B   	             
  	quantization    	Q4_0 	             
  	context length  	8192 	             
  	embedding length	3072 	             
  	                                  
  Parameters                         
  	stop            	"<start_of_turn>"	 
  	stop            	"<end_of_turn>"  	 
  	penalize_newline	false            	 
  	repeat_penalty  	1                	 
  	                                  
  License                            
  	Gemma Terms of Use              	  
  	Last modified: February 21, 2024

[–] jacksilver@lemmy.world 3 points 4 months ago (1 children)

Since there is a user acceptance policy that restricts what you can do with the model that might be considered "partially" open.

Yeah you can see the weights, but it seems you are limited on what you can do with the weights. How we've gotten to the point you can protect these random numbers that I've shared with you through a UA is beyond me.

[–] eager_eagle@lemmy.world 4 points 4 months ago (1 children)

the same happens with BloomZ, and that is listed as open

[–] jacksilver@lemmy.world 3 points 4 months ago

Wasn't aware of that, I was just taking a guess.

That being said I wouldn't consider either open given those restrictions.

[–] passepartout@feddit.org 6 points 4 months ago

It's already on Ollama. Exciting times!

[–] BaroqueInMind@lemmy.one 3 points 4 months ago* (last edited 4 months ago) (1 children)

I'm looking at the HuggingFace leader boards and Gemma isn't even top 10. How does it stack up against Llama3 or Mistral-Open-Orca?

[–] simple@lemm.ee 3 points 4 months ago

It's currently number 11 in the LMSys chatbot leaderboard. It's sitting above Llama 3 and Claude 3 Sonnet.