7 2 months ago

This model uses the followin training techniques. Please note that this is a play model for learning. It has been trained on tool use but due to its extremely small foundation model, it is unlikely to give reliable tool output. The main goal was to enable

5c0a5c816ad2 · 216B
{
"num_ctx": 1024,
"num_predict": 512,
"num_thread": 4,
"repeat_last_n": 64,
"repeat_penalty": 1.1,
"stop": [
"<|endoftext|>",
"<|user|>",
"</function_call>"
],
"temperature": 0.7,
"top_k": 50,
"top_p": 0.9
}