AZERDSQ/ g1-nano-instruct:latest

6 hours ago

A 60M-parameter GPT pretrained and instruction-tuned on a single 8GB-RAM NVIDIA Jetson, offering a native 2048-token context and real multi-turn conversation support.

... /
params
54bcc9d276ac · 63B
{
"num_ctx": 2048,
"stop": [
"<|end|>",
"</s>"
]
}