Vace powercharged

docs/FINETUNES.md

@@ -75,10 +75,10 @@ If you launch the app with the *--save-quantized* switch, WanGP will create a qu
 2) Launch WanGP *python wgp.py --save-quantized*
 3) In the configuration menu *Transformer Data Type* property choose either *BF16* of *FP16*
 4) Launch a video generation (settings used do not matter). As soon as the model is loaded, a new quantized model will be created in the **ckpts** subfolder if it doesn't already exist.
-5) To test that this works properly set the local path in the "URLs" key of the finetune definition file. For instance *URLs = ["ckpts/finetune_quanto_fp16_int8.safetensors"]*
+5) WanGP will update automatically the finetune definition file with the local path of the newly created quantized file (the list "URLs" will have an extra value such as *"ckpts/finetune_quanto_fp16_int8.safetensors"*
 6) Remove *--save-quantized*, restart WanGP and select *Scaled Int8 Quantization* in the *Transformer Model Quantization* property
 7) Launch a new generation and verify in the terminal window that the right quantized model is loaded
-8) In order to share the finetune definition file you will need to store the fine model weights in the cloud. You can upload them for instance on *Huggingface*. You can now replace in the definition file the local path by a URL (on Huggingface to get the URL of the model file click *Copy download link* when accessing the model properties)
+8) In order to share the finetune definition file you will need to store the fine model weights in the cloud. You can upload them for instance on *Huggingface*. You can now replace in the finetune definition file the local path by a URL (on Huggingface to get the URL of the model file click *Copy download link* when accessing the model properties)
 
 You need to create a quantized model specifically for *bf16* or *fp16* as they can not converted on the fly. However there is no need for a non quantized model as they can be converted on the fly while being loaded.