Of all possible prompts they shared this one.

So much potential🥹

1: One moment, let me state the vision
No more sliding windows, we on a mission!
udio.com/songs/mfsfmg5V…

2: No 1x1, that's just a view
Yann LeCun can't argue, our models slew
udio.com/songs/eb2bmgzT…

There are bound to be some rough edges with this release. Let us know as you encounter them, and we will do our best to resolve them!

This is probably the best announcement post of the day!

RecurrentGemma is close to Gemma-2B while being trained on 1T fewer tokens (!), making it one of the strongest open models at 2B scale.

+ performs very well vs much larger 7B Mistral on human evals testing instruction following and safety.

Details here: storage.googleapis.com/deepmind-media…

Announcing RecurrentGemma!
github.com/google-deepmin…

- A 2B model with open weights based on Griffin
- Replaces transformer with mix of gated linear recurrences and local attention
- Competitive with Gemma-2B on downstream evals
- Higher throughput when sampling long sequences

This is exactly what I hate with all big frameworks. TF is terrible. PyTorch used to be straightforward but turned terrible too. Torch7 was very direct. JAX/Flax still ok, but I pray every day that it doesn’t end up with the same fate over time.

OK, now I just need to convince my wife that this is a great investment of 1.7M chf. I'm sure that domain name will double in value over the next year!

something something bubble

Confirmed and useful!

I am very happy to announce that Gemma 1.1 Instruct 2B and “7B” are out! Here are a few details about the new models:
1/11