HuangChiEn
HuangChiEn
that's why i don't want to take a autonomous taxi www
add following code snippet, after `llava.train.train.py` Line 1568 : Namely, after this line : https://github.com/LLaVA-VL/LLaVA-NeXT/blob/09e5840d5589ad2d6a8656c0a60f21ae134b3309/llava/train/train.py#L1568 ``` # hack for the LLaMA 3 using Qwen tokenizer, by default Qwen doesn't have...
> I had a PR over here [#37406](https://github.com/huggingface/transformers/pull/37406) (kinda forgot about it myself). For now, you would need to update your torch to 2.6.x to avoid the compile errors you...
> Hi [@HuangChiEn](https://github.com/HuangChiEn) can you link to the code snippet you got this from? > Hi [@HuangChiEn](https://github.com/HuangChiEn) can you link to the code snippet you got this from? thanks for...
> 2.6.x oh.., **i may have a try**. > Because i may also struggled in pkg dependence, i had tried several combination to make those competible to torch==2.5 several days...
By the way, is it possible to apply `flex_attention` ? recently i just search a bit about it, seems flexible attention (flex attention) integrate most of important functionality for reducing...
> This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread....
> Flex attention works great for training atm, for inference it's a mixed bag atm especially since it depends from version to version what works with it or not -...
> Good questions. I see InternVL3.0's conv_type is also internvl2_5, so i guess conv_type = internvl2_5 is general? And btw, in internvl_chat_gpt_oss/shell there exists scripts for internvl3_5 fine-tuning. Hope this...
Actually there's one way, but now i'm too lazy to do so ~ Officially, internVL3.5 apply `apply_chat_template`, which do the same mechanism as **Conversation class**, so we can inspect the...