Update app.py

app.py

@@ -243,7 +243,6 @@ def uploadNote(prompt,num_inference_steps,guidance_scale,timestamp):
         f.write(f"Model UNET: ford442/RealVisXL_V5.0_BF16 \n")
     upload_to_ftp(filename) 
 
-@torch.no_grad()
 def captioning(img):
     prompts_array = [
        # "Adjectives describing this scene are:",
@@ -265,7 +264,7 @@ def captioning(img):
         **inputsa,
         do_sample=False,
         num_beams=5,
-        max_length=768,
+        max_length=256,
         min_length=64,
         top_p=0.9,
         repetition_penalty=1.5,
@@ -280,12 +279,13 @@ def captioning(img):
     # Loop through prompts array:
     for prompt in prompts_array:
         inputs = processor5(images=img, text=prompt, return_tensors="pt").to('cuda')
+        #with torch.no_grad():
         generated_ids = model5.generate(
             **inputs,
             do_sample=False,
             num_beams=5,
             max_length=64,
-            #min_length=42,
+            min_length=24,
             top_p=0.9,
             repetition_penalty=1.5,
             length_penalty=1.0,
@@ -297,12 +297,24 @@ def captioning(img):
         print(f"{response_text}\n")
     inputf = processor5(
         images=img,
-        text=generated_text + 'So therefore',
+        text=generated_text + ' So therefore, ',
         return_tensors="pt"
     ).to('cuda')
-    generated_ids = model5.generate(**inputf, max_length=128)
+    generated_ids = model5.generate(
+            **inputf,
+            do_sample=False,
+            num_beams=5,
+            max_length=256,
+            min_length=24,
+            top_p=0.9,
+            repetition_penalty=1.5,
+            length_penalty=1.0,
+            temperature=1,
+        )
+    )
     generated_texta = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
     response_text = generated_texta.replace(generated_text, "").strip()
+    print(f"{response_text}\n")
     output_prompt.append(response_text)
     output_prompt = " ".join(output_prompt)
     return output_prompt
@@ -437,8 +449,12 @@ def generate_30(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
+        print("-- CURRENT PROMPT --")
+        print(prompt)
         prompt = " ".join(prompt)
-        captions = " ".join(captions)
+        print("-- CURRENT PROMPT AFTER .join --")
+        print(prompt)
+        captions = " ".join(caption)
         print(captions)
         print("-- not generating further caption --")
         global model5
@@ -562,8 +578,12 @@ def generate_60(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
+        print("-- CURRENT PROMPT --")
+        print(prompt)
         prompt = " ".join(prompt)
-        captions = " ".join(captions)
+        print("-- CURRENT PROMPT AFTER .join --")
+        print(prompt)
+        captions = " ".join(caption)
         print(captions)
         print("-- not generating further caption --")
         global model5
@@ -687,8 +707,12 @@ def generate_90(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
+        print("-- CURRENT PROMPT --")
+        print(prompt)
         prompt = " ".join(prompt)
-        captions = " ".join(captions)
+        print("-- CURRENT PROMPT AFTER .join --")
+        print(prompt)
+        captions = " ".join(caption)
         print(captions)
         print("-- not generating further caption --")
         global model5