4skl/ gemma4-e2b-mtp:latest

583 1 month ago

Ultra-fast multimodal 2.3B Gemma 4 for on-device edge AI (3.7GB). Adds native vision/audio parsing to Unsloth Dynamic QAT with a 2-token MTP pipeline and a mobile-safe 32K context window. Perfect for smartphones and laptops sharing 8GB of total RAM.

vision tools thinking
... /
params
205796ff7d83 · 87B
{
"draft_num_predict": 2,
"num_ctx": 32768,
"stop": [
"<turn|>"
],
"temperature": 0.4
}