revert back to blip

cog_test_train.sh

@@ -7,6 +7,5 @@ cog predict --gpus $GPU_ID \
     -i concept_mode="face" \
     -i sd_model_version="sdxl" \
     -i max_train_steps="300" \
-    -i caption_model="florence" \
     -i debug="False" \
     -i seed="0"

predict.py

@@ -87,11 +87,6 @@ class Predictor(BasePredictor):
             description="Rank of LoRA embeddings for the unet.",
             default=16
         ),
-        caption_model: str = Input(
-            description="Which model to use for captioning the images",
-            choices=["blip", "florence"],
-            default="blip"
-        ),
         n_tokens: int = Input(
             description="How many new tokens to train (highly recommended to leave this at 2)",
             ge=1, le=4, default=3
@@ -144,7 +139,7 @@ class Predictor(BasePredictor):
             ti_lr=ti_lr,
             unet_lr=unet_lr,
             lora_rank=lora_rank,
-            caption_model=caption_model,
+            caption_model="blip",
             n_tokens=n_tokens,
             verbose=True,
             debug=debug,

requirements.txt

@@ -1,6 +1,6 @@
-torch==2.2.1
-torchaudio==2.2.1
-torchvision==0.17.1
+torch==2.1.0
+torchaudio==2.1.0
+torchvision==0.16.0
 transformers==4.38.0
 diffusers==0.26.0
 tokenizers==0.15.2