add first app.py

app.py

@@ -0,0 +1,35 @@
+from transformers import InstructBlipProcessor, InstructBlipForConditionalGeneration
+from PIL import Image
+import torch
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+print(f"Using device: {device}")
+
+# Load model (fits in 12GB with fp16)
+model_name = "Salesforce/instructblip-flan-t5-xl"
+processor = InstructBlipProcessor.from_pretrained(model_name)
+model = InstructBlipForConditionalGeneration.from_pretrained(
+    model_name, 
+    torch_dtype=torch.float16, 
+    device_map="auto"
+)
+
+# Load your image
+image = Image.open("example.jpg").convert("RGB")
+
+# Prompt to force paragraph-level description
+prompt = (
+    "Describe this image in a detailed paragraph of 5-7 sentences. "
+    "Mention setting, objects, colors, actions, background details, and possible context."
+)
+
+inputs = processor(images=image, text=prompt, return_tensors="pt").to(device)
+
+out = model.generate(
+    **inputs,
+    max_new_tokens=250,  # enough for multi-sentence
+    temperature=0.7,
+    top_p=0.9
+)
+
+print(processor.batch_decode(out, skip_special_tokens=True)[0])

requirements.txt

@@ -0,0 +1,16 @@
+# Hugging Face ecosystem
+transformers==4.41.2
+accelerate==0.33.0
+safetensors==0.4.3
+sentencepiece==0.2.0
+
+# Parameter-efficient fine-tuning (LoRA)
+peft==0.11.1
+bitsandbytes==0.43.1
+
+# Utilities
+pillow==10.3.0
+numpy==1.26.4
+
+# Demo UI
+gradio==4.37.2