Vaibhav Srivastav’s Post

Gemini

4B Parma multi model is insane!

Gemma 3n sounds wild, but I’m wondering how its multimodal performance stacks up against open models like Yi or OpenMoE. Anyone tried pushing it beyond the demo scope yet? Curious how it handles noisy inputs.

Google is 👑 data

Impressive

Very impressive light model from Google: Gemma 3n 4B/2B ! Short resume from the docs( https://ai.google.dev/gemma/docs/gemma-3n#parameters ) * Chatbot Arena Elo score: 1303 Gemma 3n (4B), 1223 Phi 4 ( 14B) ( for text only should be interesting to see the Elo Score with the quantized version like https://huggingface.co/unsloth/gemma-3n-E4B-it-GGUF/blob/main/gemma-3n-E4B-it-Q6_K.gguf) 😀. * open weights and licensed for responsible commercial use * Offline agentic use and built for Privacy, no connection required * Audio input: speech recognition, translation, and audio data analysis. * Visual and text input: Multimodal capabilities let you handle vision, sound, and text * PLE caching: Per-Layer Embedding (PLE) parameters contained in these models can be cached to fast, local storage to reduce model memory run costs. * MatFormer architecture: Matryoshka Transformer architecture allows for selective activation of the models parameters per request. * Conditional parameter loading: Bypass loading of vision and audio parameters in the model to save memory resources. * Wide language support: trained in over 140 languages. * 32K token context: Substantial input context

🧑🍳 🦙

Very cool models for solo builders!

Google is coming back for the crown! Come share what you build and learn with 5,000+ of us in the AI Agents group on linkedin: https://www.linkedin.com/groups/6672014

I'm eager to use it to enhance all kinds of products and services, making them more human surpassing that computer like feel most systems have nowadays. A new era of edge AI products is sure to arise!