metadata
library_name: peft
base_model: Qwen/Qwen3-8B
base_model_relation: adapter
language:
- bn
tags:
- lora
- peft
- bangla
- grammar-correction
- spelling-correction
- adapter-merge
- shuddho
pipeline_tag: text-generation
Shuddho Qwen3-8B 118K Merged LoRA
Merged from:
- Previous 98K LoRA adapter
- Newly trained 20K LoRA adapter
Merge method: SVD weighted adapter merge, weights=[1.0, 1.0], svd_rank=16.