laya-bangla-lora
A LoRA adapter that turns laya-multilingual into a typed-decision model without shipping full weights.
Community card
What this is
An adapter rather than a full checkpoint. You need convaiinnovations/laya-multilingual alongside it, and the licence of that base applies to you as well as this repository's own.
Its card lists training data: mteb/amazon_massive_intent, mteb/sib200, aplycaebous/BanTH, aplycaebous/BanHate, TanjimKIT/Bangla_Sentiments_in_eLearning. That matters when you read any benchmark number about it, because a model tuned on a benchmark's own training split has an advantage over one that has never seen it.
How to run it
ONNX Runtime
An ONNX export loads without PyTorch, which matters for slim containers and for browser or edge runtimes. Full ONNX Runtime guide.
pip install onnxruntime huggingface_hubA public number is a shortlist, not a decision
The only benchmark that settles which model you ship is one built from your own historical decisions. We freeze that test set, run the candidates against it, and hand back the threshold your error budget supports.