S68S がリツイート

We integrated huihui-ai/Huihui-Qwen3.5-35B-A3B-abliterated and Jackrong/Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-Distilled using MoE to create a 67B-A3B model. Currently, the tests are proceeding normally, and we are still conducting further tests.
The goal is to create a model that can both perform ablation and retain the content from Claude.
The number of experts has been increased from 256 to 512.
English





























