JXW67/ TGAI_NB:v2.5

40 1 month ago

TGAI NB — a lightweight Chinese MoE chat model family, built end-to-end by a high school student. V2 (3B): 8-expert sparse MoE, better quality for desktop. V1 (0.86B): 4-expert sparse MoE, ultra-light, smooth on low-end devices.

0.86b 3b
... /
params
1391927ba7ac · 179B
{
"num_ctx": 4096,
"num_predict": 2048,
"repeat_penalty": 1.3,
"stop": [
"用户:",
"TGAI?",
"TGA?",
"<0x",
"<EOS>",
"<BOS>"
],
"temperature": 0.3,
"top_k": 40,
"top_p": 0.9
}