Chantal Pellegrini

Results 7 comments of Chantal Pellegrini

Batch evaluation

For me, batch evaluation also gave nan values, after fine-tuning. Performing evaluation also with bfloat16 instead of float16 solved this for me. (I also fine-tuned using bf16 True). Like this,...

Radiology Q-A dataset

Sounds really interesting! What's the current status of this? :)

Radiology Q-A dataset

@hashdaddyd I cannot find your radiopaedia dataset on huggingface anymore. Is there any way to make it available again? :)

Pre-Training BLIP2 Log

Hi @phellonchen, could you maybe elaborate on what you needed to change to perform stage 1 pre-training? That would be really helpful!

Stage 1 pre-training

Thanks a lot for your answer and for publishing the pretraining code! 😊

Query Regarding Input Shape and Reproduction of Precomputed Embeddings

Hi, An image size of 448 seems to be correct, could you let me know where in the code you are facing this problem? Thanks

Query Regarding Input Shape and Reproduction of Precomputed Embeddings

Hi, I now understand where you come from. This version of RaDialog (this repo) follows a BLIP-style architecture. Therefore the embeddings are not directly the output of the biovil encoder...