Hugging Face's logo

Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
Back to feed
danielhanchen 
posted an update 18 days ago
Andyx1976

like normal 12b, 12b qat, e4b and e2b, it is completely hopeless with any vision task. Cool prove of future concept though.
johnlockejrr

Faster indeed but doesn't compare in performance with original gemma-4
evewashere

Framing tokens/sec "in step" as just tokens/sec without saying "yeah its basically the same speed as normal gemma 26ba4b because it has to go over each token 20 times" is misleading.

pavel-ai

What hardware has been used for this test?
In this post