remyxai / VQASynth

Compose multimodal datasets 🎹
https://twitter.com/smellslikeml/status/1756723056675094726
216 stars 13 forks source link

getting tis error very often: #7

Closed monjha closed 1 week ago

monjha commented 2 months ago

Getting this error very often: llava_eval_image_embed : failed to eval caption_processor-1 | Llama.generate: prefix-match hit

Tried to run for more than 24 hours and still the label file wasn't generated,

salma-remyx commented 1 week ago

Hey @monjha ! Thanks for bringing this to our attention - we've recently updated the full pipeline to use more lightweight models which should help address the issue above. The previous version of the pipeline relied on a larger model for captioning which caused it to be very slow