-
-
Notifications
You must be signed in to change notification settings - Fork 5.3k
Home
oobabooga edited this page Jun 13, 2024
·
26 revisions
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.