Gemma 2 2B IT

Gemma 2 2B IT keeps impressing AI enthusiasts with its capabilities despite its tiny size. Here's what some users have said about it:

• Those who have tried it reported that for simpler tasks, it performs better than Phi-3 Mini (and sometimes Llama 3.1 8B IT) in writing stories and following instructions. It's quite coherent in conversation, often answering like a human.

• Users found it quite good at simpler math and reasoning tasks.

• In data extraction tasks, its output is reportedly comparable to its larger sibling, Gemma 2 9B, with only slight differences.

• Some users noted that while previous small models were just for experimentation, this new model might be more practical for everyday use.

• Android users trying it for writing and roleplaying said it's faster than Phi-3. To test the model, you can use Layla App for Android. (In this post, we described the experience of users testing the new model on Android and iOS devices.)

• There was a mention that it handles Portuguese 🇵🇹 better than some much larger models (Phi 3 isn't also so good in comparison). However, its Dutch language skills were described as poor.

• AI enthusiasts suggest this might be the smallest model with good Markdown support, which they find great.

Still, as some users pointed out, being impressed by the capabilities of the model given its size doesn't necessarily mean it will be suitable for all real-world use cases 💯. So, let's wait and see how it performs in various applications!