docs/What Works.md

What Works

Loader	Loading 1 LoRA	Loading 2 or more LoRAs	Training LoRAs	Multimodal extension	Perplexity evaluation
Transformers	✅	✅**	✅*	✅	✅
llama.cpp	❌	❌	❌	❌	use llamacpp_HF
llamacpp_HF	❌	❌	❌	❌	✅
ExLlamav2_HF	✅	✅	❌	❌	✅
ExLlamav2	✅	✅	❌	❌	use ExLlamav2_HF
AutoGPTQ	✅	❌	❌	✅	✅
AutoAWQ	?	❌	?	?	✅
HQQ	?	?	?	?	✅

❌ = not implemented

✅ = implemented

* Training LoRAs with GPTQ models also works with the Transformers loader. Make sure to check "auto-devices" and "disable_exllama" before loading the model.

** Multi-LoRA in PEFT is tricky and the current implementation does not work reliably in all cases.