The new 3B "fully open source" model from AMD

Submitted ⁨⁨1⁩ ⁨year⁩ ago⁩ by ⁨foremanguy92_@lemmy.ml⁩ to ⁨technology@lemmy.world⁩

https://rocm.blogs.amd.com/artificial-intelligence/introducing-instella-3B/README.html

This is again a big win on the red team at least for me. They developed a “fully open” 3B parameters model family trained from scratch on AMD Instinct™ MI300X GPUs.

AMD is excited to announce Instella, a family of fully open state-of-the-art 3-billion-parameter language models (LMs) […]. Instella models outperform existing fully open models of similar sizes and achieve competitive performance compared to state-of-the-art open-weight models such as Llama-3.2-3B, Gemma-2-2B, and Qwen-2.5-3B […].

As shown in this image (rocm.blogs.amd.com/…/scaling_perf_instruct.png) this model outperforms current other “fully open” models, coming next to open weight only models.

A step further, thank you AMD.

PS : not doing AMD propaganda but thanks them to help and contribute to the Open Source World.

source

Comments

Sort:hotnew top

humanspiral@lemmy.ca ⁨1⁩ ⁨year⁩ ago
OpenCL not mentioned, and so raw hardware level code most likely. Maybe no one else cares, but higher level code means more portability.

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  What is the link with rocm?
  
  source
MITM0@lemmy.world ⁨1⁩ ⁨year⁩ ago
I’ll be bookmarking the website & thank you

source
Ulrich@feddit.org ⁨1⁩ ⁨year⁩ ago
I don’t know why open source malicious software is worthy of praise but okay.

source
- domi@lemmy.secnd.me ⁨1⁩ ⁨year⁩ ago
  I’ll bite, what is malicious about this?
  
  source
  - Ulrich@feddit.org ⁨1⁩ ⁨year⁩ ago
    What’s malicious about AI and LLMs? Have you been living under a rock?
    
    At best it is useless, and at worst it is detrimental to society.
    
    source
    -> View More Comments
MonkderVierte@lemmy.ml ⁨1⁩ ⁨year⁩ ago
It’s about AI.

source
1rre@discuss.tchncs.de ⁨1⁩ ⁨year⁩ ago
Every AI model outperforms every other model in the same weight class when you cherry pick the metrics… Although it’s always good to have more to choose from

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  I’ve shared this AI because it’s one of the best fully open source AI
  
  source
art@lemmy.world ⁨1⁩ ⁨year⁩ ago
Help me understand how this is Open Source? Perhaps I’m missing something, but this is Source Available.

source
- frezik@midwest.social ⁨1⁩ ⁨year⁩ ago
  The source code on these models is almost too boring to care about. Training data and weights is what really matters.
  
  source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  Instead of the traditional open models (like llama, qwen, gemma…) that are only open weight, this model says that it has :
  
  Fully open-source release of model weights, training hyperparameters, datasets, and code
  
  Making it different from other big tech “open” models. Tough it exists other “fully open” models like GPT neo, and more
  
  source
BitsAndBites@lemmy.world ⁨1⁩ ⁨year⁩ ago
Nice. Where do I find the memory requirements? I have an older 6GB GPU so I’ve been able to play around with some models in the past.

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  Following this page it should be enough based on the requirements of qwen2.5-3B qwen-ai.com/requirements/
  
  source
- ikidd@lemmy.world ⁨1⁩ ⁨year⁩ ago
  LMstudio usually lists the memory recommendations for the model.
  
  source
- Danitos@reddthat.com ⁨1⁩ ⁨year⁩ ago
  No direct answer here, but my tests with models from HuggingFace measured about 1.25GB of VRAM per 1B parameters.
  
  Your GPU should be fine if you want to play around.
  
  source
A_A@lemmy.world ⁨1⁩ ⁨year⁩ ago
Nice and open source . Similar performance to Qwen 2.5.
(also … tomsguide.com/…/i-tested-deepseek-vs-qwen-2-5-wit… ← tested DeepSeek vs Qwen 2.5 … )
→ Qwen 2.5 is better than DeepSeek.
So, looks good.

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  Dont know if this test in a good representation of the two AI, but in this case it seems pretty promising, the only thing missing is a high parameters model
  
  source
Zarxrax@lemmy.world ⁨1⁩ ⁨year⁩ ago
And we are still waiting on the day when these models can actually be run on AMD GPUs without jumping through hoops.

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  That is a improvement, if the model is properly trained with rocm it should be able to run on amd GPU easier
  
  source
- grue@lemmy.world ⁨1⁩ ⁨year⁩ ago
  In other words, waiting for the day when antitrust law is properly applied against Nvidia’s monopolization of CUDA.
  
  source
Canadian_Cabinet@lemmy.ca ⁨1⁩ ⁨year⁩ ago
I know it’s not the point of the article but man that ai generated image looks bad. Like who approved that?

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  Oh yeah you’re right :-)
  
  source
SnotFlickerman@lemmy.blahaj.zone ⁨1⁩ ⁨year⁩ ago

3B

That’s one more than 2B so she must be really hot!

/nierjokes

AMD knew what they were doing.

source
- altkey@lemmy.dbzer0.com ⁨1⁩ ⁨year⁩ ago
  Can’t judge you for wanting to **** her or whatever, just don’t ask her for freebies. She won’t care if you are a human at that point.
  
  source
- greybeard@lemmy.one ⁨1⁩ ⁨year⁩ ago
  That’s a real stretch. 3B is basically stating the size of the model, not the name of the model.
  
  source
  - weew@lemmy.ca ⁨1⁩ ⁨year⁩ ago
    Are you calling her fat?
    
    source
    -> View More Comments
TheGrandNagus@lemmy.world ⁨1⁩ ⁨year⁩ ago
Properly open source.

The model, the weighting, the dataset, etc. every part of this seems to be open. One of the very few models that comply with the Open Software Initiative’s definition of open source AI.

source
- foremanguy92_@lemmy.ml ⁨1⁩ ⁨year⁩ ago
  Look at the picture in my post.
  
  There was others open models but they were very below the “fake” open source models like Gemma or Llama, but Instella is almost to the same level, great improvement
  
  source