tranhuythang9999/ deepseek-v4-pro-9b-heretic:probe

165 yesterday

9B distilled from DeepSeek-V4-Pro: decensored, with thinking, tool calling and vision in one checkpoint. q4km 6.25GB · q6k 7.98GB · q8_0 10.15GB · f16 18.84GB.

vision tools thinking
... /
params
825ff9c7727d · 158B
{
"num_ctx": 32768,
"num_predict": 4096,
"repeat_penalty": 1.05,
"stop": [
"<|im_start|>",
"<|im_end|>"
],
"temperature": 0.6,
"top_k": 20,
"top_p": 0.95
}