38 yesterday

Gemma-4 26B A4B Agentic SFT, high convergence and virtually no looping

vision tools thinking
318207667fee ยท 413B
{
"draft_num_predict": 2,
"min_p": 0,
"presence_penalty": 0.025,
"repeat_penalty": 1,
"temperature": 0.8,
"think_budget": "high",
"think_budget_message": "\n\nI have used my thinking budget. I must stop analysing now and act on what I have: make the tool call, or give a short final answer if no call is needed.\nI'll be more terse and concise now and maybe I need to consider a different approach.\n",
"top_k": 64,
"top_p": 0.95
}