bjoernb/
deepseek-r1-8b:latest

54 1 month ago

DeepSeek-R1 is a family of open reasoning models with performance approaching that of leading models, such as O3 and Gemini 2.5 Pro.

tools thinking
... /
params
34e76c8ecbc5 · 203B
{
"num_predict": 16384,
"repeat_penalty": 1.1,
"seed": -1,
"stop": [
"REDACTED_SPECIAL_TOKEN",
"REDACTED_SPECIAL_TOKEN",
"REDACTED_SPECIAL_TOKEN",
"REDACTED_SPECIAL_TOKEN"
],
"temperature": 0.2,
"top_k": 10,
"top_p": 0.9
}