Gemma 4 12B: A unified, encoder-free multimodal model
1–10 of 421 posts
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#2The big story here is the encoder-free part, which I still don't fully understand.
> Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations.
That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is robust enough. https://developers.googleblog.com/gemma-4-12b-the-developer-...
> Small enough to run locally on consumer laptops with 16GB of RAM, it unlocks powerful multimodal and agentic experiences right on your machine.
I am assuming that involves quantization, which due to the quality loss makes that statement somewhat misleading IMO.
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#3Wow Google is becoming the new pre Llama 4 Meta when it comes to releasing open weights models.
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#4The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
> That's technically encoding
Isn't that just projecting the patches into the d_model size vectors that the models takes?
>I am assuming that involves of quantization
12B model in 16GB seems very reasonable to me, int8 is top quality for running models.
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#5The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
> quantization
12b means 12G @ 8 bits/param (basically lossless) and 6G at 4 b/p (generally accepted 'pretty close' level). Not too bad?
But TBD how well the base model performs before thinking too much about quantization
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#6The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
Totally agree that it is "encoding" in the general sense, but I think they are referring to the lack of an "encoder" neural network.
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#7The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
It actually works well because unlike encoders, the latent space is trained on that initial layer so it “knows” what to do with that sparse density. I’ve been using gemma4-12b with Flux2 and its ability to reason on visual input is pretty good. That said, each model is good in their own ways so YMMV but overall, it’s about as solid as Qwen just with a more advanced architecture.
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#8The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
[dead]
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#9The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
Well its a real simple encoder I guess
Re: Gemma 4 12B: A unified, encoder-free multimodal model
#10The big story here is the encoder-free part, which I still don't fully understand. > Vision: We replaced Gemma 4’s vision encoder with a lightweight embedding module consisting of a single matrix multiplication, positional embedding and normalizations. That's technically encoding, just without using a dedicated model for it like SigLIP? The Developer's Guide elaborates, it's still a 35M layer which I am curious is ro…
Totally agree that it is "encoding" in the general sense, but I think they are referring to the lack of an "encoder" neural network.
In hindsight I may have been pedantic.