ps, it's just a hobby, I'm not from any AI Labs, just a noob, so don't hesitate to give those suggestion even if it seems dumb and obvious for Andrej Kaprathy level researcher.
ps.2, I'm from UTC+8, so I might not be able to reply for next 10 hours or so.
(when I say part of, to be more specific, it's context_embedder condition_proj and token_refiner, and using a cross-attention encoder to transform Qwen3VL's vocab list into Qwen3.5's vocab list )
I hope I can get some of your prompt to further train this model. I also afraid of overfitting. any suggestion about how to do so more efficiently and source of prompt would be good!
I'm wondering is there anyone who truly believe Mistral is going to release a model called Le Chaton Fat. As far as I know, it's just a meme. But I saw somebody is "waiting is technical report", I'm quite confused
