44 yesterday

A lightweight local intent-analysis model built for the OpenViking search API. It detects whether a conversation turn needs context retrieval, skips chitchat to reduce unnecessary memory injection and token usage, and emits structured retrieval queries

tools thinking
e507fe24898d · 85B
{
"stop": [
"<|im_end|>",
"<|im_start|>",
"<|endoftext|>"
]
}