Colab compatibility after Gemma 4 patches?

[MetaMind42]

Is the 26B A4B notebook from Unsloth still functionally relevant after all the Gemma 4 patches?

[pauldev-hub]

Hi @MetaMind42 ,

Yes — the notebook is still relevant and fully functional, but you need the latest Unsloth version.
Unsloth trains Gemma 4 ~1.5x faster with ~60% less VRAM. Studio and the notebooks work for Vision, Text, Audio and inference for 26B-A4B. (GitHub)

The Gemma 4 patches fixed real bugs that were NOT from Unsloth:

Gradient accumulation was causing loss explosions (300–400 instead of 10–15) — fixed
IndexError on 26B/31B inference — fixed
use_cache=False gibberish on E2B/E4B — fixed
GGUFs re-converted — you need to re-download if you had old ones

These quant issues are not related to or caused by Unsloth. The fixes are now included in both Unsloth and Unsloth Studio. (Unsloth AI)

What you need to do before running the notebook:
pip install --upgrade unsloth unsloth_zoo
Also note: Do NOT use CUDA 13.2 runtime for any GGUF as it will cause poor outputs.

If this helped you, please mark it as the answer — it helps others in the community who run into the same issue find the solution faster!