Specialized 3B Models Now Outperform Frontier APIs on Enterprise OCR Tasks at 50x Lower Cost
Dharma's DharmaOCR benchmark shows task-specific fine-tuning can beat parameter scale in production AI economics.
#fine-tuning
bitsandbytes: The Open-Source Engine Behind Accessible LLM Fine-Tuning
The bitsandbytes library applies 4-bit and 8-bit quantization to PyTorch models, making 70B+ parameter LLMs runnable on consumer GPUs and underpinning the QLoRA fine-tuning wave.