AZERDSQ/ g1-nano-base:latest

7 hours ago

A 60M-parameter GPT trained from scratch on a single 8GB-RAM NVIDIA Jetson, offering a native 2048-token context for raw text completion. Base pretrained checkpoint, not instruction-tuned.

... /
params
183781e216dc · 43B
{
"num_ctx": 2048,
"stop": [
"</s>"
]
}